Future Trends in Biomedical and Health Informatics and Cybersecurity in Medical Devices

Until today, In the daily work of dental radiologists, the flipping and identification of the intraoral x-ray images must be controlled by humans. No software or artificial design software can be improved, which causes the radiologist to take time and flip the teeth and put them into the correct tooth area. Therefore, under the increasing workload, how to shorten the time spent by the radiologist in handling is particularly important. In this study, through the Convolutional Neural Network (CNN) architecture in Deep Learning, the most commonly used intraoral films (Periapical, Horizontal Bite Wing, Vertical BiteWing) were inverted and identified to show the tooth area, and 16 tooth positions were designed for AI identification and learning; a total of 15,752 dental x-ray films (including original images, inverted 90°, 180° and 270° each 3938) for AI training. Therefore, in this study, 328 tests were performed after AI training (testing 1. tooth position recognition 2. flipping 90°, 180°, 270° image recognition each time), the recognition success rate is: 1. tooth position identification the success rate was 97.56%. 2. The average success rate of flip image recognition was 97.56%. Therefore, there is a great relationship between the number of AI learning images and the classification of image features and the success rate of recognition. In this study, the AI has been linked to the browser, and it can be used for teaching and research in the industry, research reference, etc., and is expected to be used clinically to reduce the workload of dental radiologists and dentists.

Minimal invasive device insertion into the soft tissue is a very important medical approach that is helpful for local drug delivery, biopsy, regional anesthesia, blood sampling, and radiofrequency ablation. These methods need to introduce the MID inside the body for exploring the tissue texture pattern. We propose a novel approach by acquiring acoustic emission data from the proximal end of the MID for the tissue using phase angle, power spectral to identify the texture. By performing these signal processing approach to the soft tissue, the pattern of the tissue can be understood. The result shows a new study to obtain soft tissue texture information.

Medical facilities and technologies have been greatly improved through the application of biosensors, healthcare systems, health diagnosis and disease prevention technologies. However, wireless transmission and deep learning neural network are essential applications and new methods in biomedical engineering nowadays. Hence, authors established a new real-time and intelligent healthcare system that will help the physician’s diagnosis over the patient’s condition and will have a great contribution to medical research. Physiological conditions can be monitored and primary diagnosis will be determined which will help people primarily for personal health care. This paper focused on the collection, transmission, and analysis of physiological signals captured from biosensors with the application of deep learning and segmented analysis for the prediction of heart diseases. Biosensors employed were non-invasive composed of infrared body temperature sensor (MLX90614), heart rate and blood oxygen sensor (MAX30100) and ECG sensor (AD8232). This research used these biosensors to collect signals integrated with Arduino UNO as a central module to process and analyze those signals. ESP8266 Wi-Fi microchip was used to transmit digitized result signals to the database for deep learning analysis. The first segment of deep learning analysis is the Long-Short Term Memory (LSTM) network applied for the temperature, heart rate and arterial oxygen saturation prediction. A rolled training technique was used to provide accurate predictions in this segment. The second segment used was the Convolutional Neural Network (CNN), which comprises three hidden layers to analyze the ECG signals from the image datasets. Deep learning tools used were the powerful python language, python based Anaconda, Google’s TensorFlow and open source neural network library Keras. The algorithm was used for evaluation using the available MIT-BIH ECG database from Physionet databases which attained 99.05% accuracy and arrived at only 4.96% loss rate after 30 training steps. The implementation of the system is comprised of physiological parameter sensing system, the wireless transmission system and the deep neural network prediction system. User interfaces were also developed such as the LCD display which shows values of body temperature, heart rate and arterial oxygen saturation level. Web page and app were created to allow users or doctors for visual presentations of the results of analysis. The webpage contains information about the system, deep learning networks used, biosensors and the historical graph of about the patient’s body temperature, heart rate and oxygen saturation. It also indicates the normal ranges of the physiological parameters.

Students playing Internet game become a daily activity in campus life. They usually struggle for fun and stay at audio-visual stimuli of Internet games for a long period of time. Some players hardly controlling the impulses of game engagement despite negative consequences can be described as a term of Internet gaming disorder (IGD). They always perform a psychological property, called tolerance symptom, for increasing the amounts of playing time to achieve satisfaction. Studies found that breathing exercise can alleviate IGD symptoms because breathing can facilitate the psychophysiological regulation. Few studies, moreover, observed the effect of breathing exercise on tolerance response of IGD symptom. This study explores the prolonged effect on respiratory sinus arrhythmia (RSA) of individuals with IGD from rest to watching game videos as stimuli through abdominal breathing (AB) training. 7 persons of high-risk IGD (HIGD) and 17 persons of low-risk IGD (LIGD) were recruited. The results showed that both of HIGD and LIGD presented an increasing RSA value with AB training from rest to stimuli. In contrast to those with LIGD, those with HIGD showed higher RSA value during negative stimuli. Our findings suggested that AB training can be a potential method to reduce psychophysiological responses of persons with HIGD during game-related cue stimuli, negative game especially. It may provide researchers insight into the effect of breathing exercises on psychophysiology responses of persons with IGD and further develop related application, such as alleviative method. A further study could investigate the effect on autonomic nervous system activities for a long-period AB training.

The detection and evaluation of the shape of liver from abdominal computed tomography (CT) images are fundamental tasks in computer-assisted liver surgery planning such as radiation therapy. The contour of spleen is also a significant factor highly related to liver diseases. However, automatic and accurate liver segmentation still remains many challenges to be solve, such as ambiguous boundaries, heterogeneous appearances and highly varied shapes of the liver and spleen. To address these difficulties, we developed an automatic segmentation model based on multi-channel U-net network. Some preprocessing steps were done to elevate the performance first. Also, an approximate liver and spleen map was generated by calculating the gradient of CT images. The area which have high possibility to be liver and spleen would be select as the training set to make sure the balance of data. Then, a deep learning U-net structure was applied for the processed training data. Finally, some post-processing methods, which include k-means clustering and morphology algorithms, would be applied in our protocol. The results indicated that a high structure similarity index (SSIM) and dice score coefficient of liver and spleen segmentation model can be achieved, which were 0.9731 and 0.9508 respectively, demonstrating the potential clinical applicability of the proposed approach.

Classification of breast cancer malignancy using digital mammograms remains a difficult task in breast cancer diagnosis and plays a key role in early detection of breast cancer. Inspired by rapid progress in the field of artificial intelligence, we explored several machine learning mechanisms, i.e., Support Vector Machine (SVM), Logistic Regression, Decision Tree, Random Forest, and Deep Neural Network (DNN) given in TensorFlow and Keras deep learning frameworks, and used Python programming to predict if a patient case is malignant or benign. This retrospective study was based on the Breast Cancer Wisconsin (Diagnostic) Data Set that contains a set of 30 features, e.g., radius mean, texture mean, perimeter mean, etc., previously extracted from digital mammograms. In addition, breast cancer diagnosis results were provided as the gold standard for training and testing of the machine learning mechanisms. Based on verification results on the test set, the five machine learning mechanisms achieved the sensitivity of 94.4%, 94.4%, 91.9%, 90.8%, 98.5%, and the specificity of 92.7%, 90.5%, 92.3%, 94.6%, 91.1%, respectively. In conclusion, our machine learning mechanisms achieved competitive performance results with the state-of-art techniques presented by other researchers and may provide useful second opinion to radiologists in breast cancer diagnosis.

A high number of slip-and-fall incidents result in common injuries of daily life. The design of outsole tread pattern is one of the key factors which had a direct impact on slip resistance performance. The application of numerical simulation is an opportunity for footwear industry to evaluate the multiple outsole tread pattern design and ground condition in a more controlled and efficient manner than mechanical testing in the developing process.A complex three-dimensional (3D) FE model of the shoe was developed to evaluate the effect of outsole tread pattern design on slip resistance performance during the gait motion. The dynamic plantar pressure distributions were automatically applied as the loading condition in FE model which allowed to interpret the individualized subject condition.The herringbone tread design and higher real contact area between shoe and ground could achieve better slip resistance performance. The process of this study demonstrates the potential of numerical simulation for evaluating slip resistance performance.

Hallux valgus (HV) was one of the most frequent human foot deformities. The aim of this study was to evaluate mechanical responses and stabilities of the plate implants and Kirschner wire (KW) after the distal metatarsal osteotomy in HV treatment by using finite element (FE) analysis. A three-dimensional FE model of lower extremity was developed to evaluate the four plate fixation methods and three KW fixation methods in weight bearing. The results showed that all the plate fixations revealed better first metatarsal stability than KW. For the result of the contact pressure, the 6-holes-6-screws dynamic compression plate implant had highest result than others plate implant. Adding the bandage to the KW fixation had a highest result in all implant.

In this study, a simple, novel and inexpensive third generation electrochemical uric acid (UA) biosensor based on three-electrode point-of-care test strip was proposed. On the low-cost screen-printed three silver-based electrodes test strip, the Ag/AgCl reference electrode was first formed by simple electrodeposition with one silver electrode, then the working electrode was prepared by the successive coating of redox mediator, polymer hydrogel, and uricase onto the surface of another silver electrode, the third bare silver electrode was served as the counter electrode to complete the uric acid test strip, which has the advantages of lower oxidation potential, faster response time, higher sensitivity, and wider detecting range. The three-electrode uric acid test strip specifically and directly senses uric acid in the test sample, and the signal is promptly transferred to the amperometric readout circuit that a stable bias voltage is applied by a bandgap circuit. After initial processing, the analog signal is converted to a digital signal which is calculated with the algorithm of microcontroller to produce a value of user-readable mode. Finally, the digital result is displayed on the liquid-crystal-display (LCD) panel. The complete set of uric acid point-of-care testing system will be used to assess the uric acid condition in the urine of a urolithiasis patient and will be helpful in the diagnosis of urolithiasis.This uric acid biosensor can be used for testing uric acid either in urine or blood specimen. As combined with other biosensors such as calcium ions, pH, and conductivity, etc., it can be extended to develop a multi-parameter detection system and apply for the prevention of urolithiasis recurrence.

With the rapid development of science and technology in recent years, medical equipment is gradually being implanted in human bodies to allow patients to lead normal lives. Implantable medical devices must function in the body for long periods of time; hence, efficient rechargeable power sources must be developed to non-intrusively recharge these devices. In the charging process, implantable medical devices will damage the surrounding tissue due to the resulting heat generation. In order to enhance the charging efficiency and decrease the temperature variation, a 200-mAh Li-polymer battery was charged by a multistage sinusoidal current with the minimum-ac impedance, and an optimal rapid-charging pattern was identified by the Taguchi method. Experiment results showed that in terms of charging efficiency and battery and tissue temperature control, a multistage sinusoidal current with the minimum-ac-impedance frequency performed best compared with constant current (CC), pulse current, and constant current constant-voltage (CC-CV) charging strategies. In terms of battery temperature variation, compared to the above-mentioned three charging methods, the multistage sinusoidal current method are improved charging temperatures about 1.34 °C, 1.53 °C, and 1.71 °C, respectively; in respect to charging efficiency, efficiency improved about 5.52%, 17.4%, and 5.37%, respectively.

The spine-related disease is one of the most common musculoskeletal-related disorder in the world. Although computed tomography (CT) is an outstanding tool for investigating spinal pathology in clinical protocol, the overexposure to radiation dose issue cannot be underestimated. Therefore, the bi-planar EOS X-ray imaging was adopted as the scanning technology, which can capture the anteroposterior (AP) and lateral (LAT) view X-ray images simultaneously with ultra-low radiation doses. High quality and high contrast bi-planar X-ray images would be acquired from the EOS system and these two radiographs enable a precise three-dimensional reconstruction of vertebrae, pelvis and other parts of the skeletal system. To overcome the time-consuming issue of spine reconstruction using the EOS system, a convolution neural network (CNN) was applied to reconstruct the entire spine model. Nowadays, the CNN model has already been adopted in the transformation from 2D image to 3D scenes. Our approach represents a potential alternative for EOS reconstruction while still maintaining a clinically acceptable diagnostic accuracy.

Pedicle screws might be backed out after screw insertion. Past studies had evaluated the pedicle screws with full insertion and back-out using experimental approaches. Unfortunately, there is rare study to investigate this problem using numerical approaches. Thus, the purpose of this study was to analyze the pullout performance of spinal pedicle screws with fully inserted setting or backed out using finite element method.Twelve types of spinal pedicle screws were developed using SolidWorks. Each screw with the full insertion, backed-out 90o, and backed-out 180o were considered to evaluate their pullout performance using ANSYS Workbench. Additionally, a bone compaction technique was developed and applied in the present study.The results showed that the pullout performance of the conical pedicle screws was significantly reduced compared to that of the cylindrical pedicle screws in situation of screw back-out. Both the screw geometry and bone compaction effect were the key parameters for the evaluation of pullout performance.

Ultrasound (US) imaging is a non-invasion and non-radiation medical imaging system. One of them, the single-element transducer has significant potential in the medical device. As compared to an array transducer, it could reduce the size of the imaging system. For example, the ultrasound needle for epidural anesthesia and eyeball.In this study, we develop a free-hand US system on skin imaging. Using the 20 MHz high-frequency single element ultrasound transducer on the skin scanning. The high-frequency ultrasound could provide an excellent resolution to 0.1 mm. The bandwidth of the transducer is 20%, and the Insertion loss (IL) is –31.9 dB, and the Electromechanical coupling factor (Kt) is 0.72. Meanwhile, the free-hand apparatus which based on the optical tracking sensor is designed and applied in this system. It is convenient for scanning the skin surface.This control and display panel is designed by the LabVIEW software. The in-board filter and Hilbert transformation are used to eliminate the environment noise. To solve the problem of unstable scanning speed, the interpolation method with one order is used to fit and smooth the image. Therefore, the high-quality ultrasound images could be applied in the skin scanning.

Obstructive sleep apnea (OSA) is a chronic breathing disorder that most of the time people are oblivious of the symptoms which may delay the diagnosis and may lead to long-term health consequences such as cardiovascular and cerebrovascular diseases. Ultrasonography is currently used to discern the real behavior of the upper airway (UA) in patients with OSA. However, previous methods were not enough to reveal the possible pathophysiology and biomechanics of the human UA. The aim of this study is to use the modified optical flow (OF)-based method in tracking the dynamic tongue base motion, utilizing nine tracking points, to effectively classify which group each subject belongs to. The classification groups are normal, mild, moderate, and severe OSA. A total of 82 participants were enrolled in this study. All of them had their B-mode ultrasound image sequences obtained for 10 s. The first 5 s was recorded during eupneic breathing, and the latter part was during the performance of the Müller Maneuver (MM), a simulation of the collapse of the UA while inducing negative pressure. The results demonstrate that the four classifications are significantly different (p < 0.05). The normal group has the largest displacement, while the severe OSA group has the smallest. The normal group has the smallest tongue base area (TBA), while the severe OSA group has the largest. Both instances were also observed during the MM. Tongue area measurement during the eupneic breathing for the four groups are 18.63 ± 2.595, 20.25 ± 2.366, 20.34 ± 3.207, and 21.75 ± 2.764, respectively. During the MM, the measurements were 18.54 ± 2.701, 20.16 ± 2.428, 20.32 ± 3.190, and 21.78 ± 2.820, respectively. Noninvasive sonographic evaluation using dynamic tongue motion tracking and tongue area measurements provides quantitative assessments that can be used by the clinician to indicate individualized treatment plan for each OSA patients.

Digital tomosynthesis (DT) reduces tissue overlap and provides tomographic images of high quality with clinically acceptable low radiation dose, it begins to be recognized as an essential diagnostic tool. The scanning direction and model are more flexible in DT. However, scanning parameters of DT are not easy to be determined. In this study, we investigated the effect of dual direction scanning on artifact improvement and the optimization of scanning parameters using the INER Prototype Tomosynthesis scanner. The line-pair shape phantom with 2, 3, 5, 8, 10, 15 mm line widths and 5 mm thickness was used. The projections were acquired with 31 views over a 15° angular range in HF (Head-Foot) direction and RL (Right-left) direction, respectively. 3D images were reconstructed with ML-EM algorithm to evaluate the spread range of ghost artifact with various sweep methods (HF, RL and Dual) and object sizes. Moreover, the projection number ratios (PNR) in dual directions were also evaluated for the influence on artifacts. Under single direction sweep (HF or RL), the spread range was wide when the sweep direction paralleled with the shape direction of object (SDO); the spread range was narrow when the sweep direction is perpendicular to the SDO. The spread range of dual scan was unaffected by the SDO. The PNR in dual directions revealed a similar trend to single direction sweep when the ratio is not equal to one. Based on the above experimental results, we proposed a novel multi-direction adjustment strategy for body tomosynthesis. For the default scanning, the user should choose the isotropic dual direction scan (PNR = 1). After preview, the advanced process can use different PNR depending on the anatomy of interest to improve diagnostic image quality to reduce ghost artifact and distortion.

Hypertension is an important risk factor for stroke and cardiovascular diseases. Ambulatory blood pressure (ABP) measurement is used to estimate the continuous blood pressure. However, the cuff method ABP must have a cuff setting around the upper arm and occluding the arm’s blood circulation during the recording period, which makes some of inconveniences, including feel uncomfortable and affects the quality of sleep. The cuff-less method of ABP measurement based on the Pulse Transit Time (PTT) with electrocardiogram (ECG) and photoplethysmogram (PPG) has solved the limitation and presented potential healthcare applications. This study applies five different blood pressure regression models with the major parameter (PTT) and minor parameters (heart rate, pulse wave interval and pulse width) for estimating continuous blood pressure by regression analysis. MIMIC II clinical database is used by the correlation and consistency analysis for different blood pressure models to compare the similarity of the real and estimated blood pressure variation. The best model among the applied blood pressure models is $$ \text{PTT}_{{\text{ALL}}} - \text{BP} $$ that can perform the average correlation in 0.87 and the average RRratio in 0.68. The blood pressure regression model of $$ \text{PTT}_{{\text{ALL}}} - \text{BP} $$ provides a successful analysis model for the estimation of long-term monitoring blood pressure trend. monitor for the real and estimated blood pressure have the same trend.

Urolithiasis is a common disease with high recurrence rate. According to the record, the prevalence rate of urolithiasis is about 4–15% in Asia, Europe and America while the recurrence rate of urolithiasis is more than 50% after the treatment. The research done by the University of Chicago indicates that the presence of crystals in urine is an important factor of stone formation. Previous studies also conclude that the compound of crystals and stones are highly correlated. Therefore, it is important to analyze urine crystals accurately to prevent potential stone formation. Our Raman spectroscopic urine crystal detection system features the specially designed microfluidic chip with a chamber, the patented technology of crystal collection using Fe3O4 nanoclusters and the 785-nm excitation wavelength automatic Raman microscope we constructed. This system can instantly extract crystals from urine, and the composition of crystals are determined accurately by Raman spectroscopy. It will be a much more powerful and more efficient tool on urine crystal analysis.The clinical significance study of urine crystals on urolithiasis management mainly focuses on the relationship between morphology including auto-fluorescence of urine crystals and urolithiasis. The urine samples and renal calculi from urolithiasis patients were collected from Taipei Veterans General Hospital and Taipei City Hospital and all the samples were analysized with our Raman spectroscopic urine crystal detection system. Statistically, the pre-surgery urine samples of urolithiasis patients tend to have more atypical crystals in shape, composition and auto-fluorescence incidence than non-patients’. Our data indicates that the pattern of urine crystals plays a significant role in clinical urolithiasis management.

In this paper, we design a method to use smart phone to detect when fall accident happened, it can inform outside people or organization automatically to get help from them. The smart phone has implemented several sensors, such as the tri-axial accelerometer, electronic compass, global positioning system (GPS) etc. We will use those sensors to do fall detection.This detection system is used by placing in the waist pocket. Because people activity can be recorded real time by center of gravity in the body. We collect data for normal movements and fall events to setup a database, and then do the data analysis real time to identify if it is normal movement or fall event.To offload system operation loading and increase the efficiency on the fall detection, this paper will be divided into two parts. In the first part, in order to make the identification more lightweight, the collected data will be used to do the data blurring by weighted moving average. This way can make system easy to comparison and will not lose fall event feature. In the second part, we input the processed data to do fall detection. Three features weightlessness, impact and stillness are used to identify if it is fall event or not. If fall accident is true, system will send warning message and location automatically to the people or organization who we predefined in the system to get help.The results of our research can be used by everyone and everywhere if wireless network connection is valid. This system can be used in various environments and it is very convenient to hand carry and easy to use. So the acceptance from general people should be very high for them can use this system to get help in time to save their life or mitigate the damage.

Lung Cancer is a leading cause of death worldwide, and about 85% of lung cancer is non-small cell lung cancer (NSCLC). The staging of lymph nodes in NSCLC patients is extremely important because respective stages require different treatments. FDG-PET/CT is a gold standard for lymph node metastasis staging of NSCLC. However, the results of discriminating lymph node staging on 18F-2-fluoro-2-deoxy-d-glucose (FDG) positron emission tomography (PET)/computed tomography (CT) still needs improvement. In addition to the traditional image parameters of FDG-PET/CT such as standardized uptake value (SUV), there are many other parameters available from FDG-PET/CT images, for example, the lymphatic drainage pathway. For the purpose of a better accuracy on lymph node metastasis diagnosis on NSCLC patient in FDG-PET/CT, this research developed a computer-aided diagnosis (CAD) system to improve the diagnostic efficiency, which achieved an accuracy of 85%, a sensitivity of 82% and a specificity of 85%.

Due to population aging, early diagnosis of neurodegenerative diseases have been noticed. To fulfill such clinical need, developing a hi-performance SPECT scanner for brain function imaging was raised. In this research, a practical scanner geometry was provided and the resolution performance of imaging detectors for composing the scanner was also studied. Consider the resolution and practicality, a cylindrical scanner design, with a detector ring of 48-cm diameter and a rotatable collimator cylinder of 32-cm diameter was chosen for further development. With 1-mm pinholes placing on the collimator cylinder, a FOV (field-of-view) of 21-cm diameter and 15-cm height, also a magnification factor of 0.48 are formed. In such design, the resolution is derived by the intrinsic resolution of imaging detectors. In the aspect of imaging detector unit, three different pixel size were studied. Three GAGG detector units with pixel size of 1.8, 1.5 and 1.2 mm were built and tested. The resultant 2D crystal maps and pixelated energy spectrum were examined to see how crystal pixels being resolved. The outward appearance of the detector unit showed that no peripheral dead space exhibits and thus allows 2D scalable to achieve the scanner building need. The 2D maps of the three detector units all showed successfully distinguished crystal arrays, therefore 1.3 mm resolution for the imaging detector can be achieved at current stage. It means the best scanner resolution at the FOV center of 2.88 mm is expected. In this study, a practical scanner geometry design is made, also its imaging detector unit is developed. Preliminary results show that the best resolution performance is better than 3 mm. Therefore the following task is to design the pinhole collimator pattern, trying to maximize the scanner sensitivity while keeping the resolution around 3 mm.

This paper presents the design of an analog front-end of an electrochemical conductivity sensor. This forms part of a multi-parameter sensing device for use in the detection and diagnosis of urolithiasis in its preliminary stages. The proposed conductivity bio-sensor is used for measuring the concentration of total dissolved salts in urine. It will be combined with FET-based potentiometric sensors for measuring Ca2+ and pH, and an amperometric sensor for measuring uric acid into an integrated lab-on-a-chip point-of-care device. Furthermore, measured data are collected and stored in a database for future use in an AI (Artificial Intelligence) diagnosis system. Such a multi-parameter approach caters for a more effective stone-risk indexing.The conductivity sensor readout circuit works on the principle of electrochemical impedance spectroscopy. The sample under test (urine) is subjected to a constant sinusoidal current which causes it to develop a potential difference. The sample interface is via four-point gold coated flat electrodes. The true RMS (root mean square) value of current and voltage is measured and impedance magnitude is calculated by dividing voltage to current.Measurement is iterated for several frequencies in the range between 1.5 kHz and 3.3 kHz. The frequency is selected from prime numbers to avoid harmonics distortion.The proposed system has been tested in an on-board platform and the results of the measurement are correlated to a commercial conductivity meter device. The developed conductivity sensor and its readout circuitry have potential usage in point-of-care test device for urolithiasis prognosis and screening.

Paroxysmal AF (PAF) is a form of atrial fibrillation (AF) that is generally clinically silent and undetected. AF is a type of heart disease called cardiac arrhythmia. Automatic detection of AF could make a significant contribution to early diagnosis, control and prevention of chronic AF complications. In this paper, authors presented a novel algorithm through spectrogram and deep learning neural network analysis in detecting paroxysmal AF from image data segments. This method does not require the detection of P and/or R peaks which is a preprocessing step required by many existing algorithms. The PAF Prediction Challenge Database from Physionet.org were used as learning set which composed of 50 record sets. These records were converted into 7,000 PAF and 964 healthy data segments. Each data segment has 5 mins-duration and converted it to graph images. These graph images are then converted into spectrogram to visualize the frequency band present in the spectrum. In this process, ECG numerical values were interpreted into spectrogram form. Spectrogram images are cropped to remove unnecessary markings from the graphing and spectrogram processes. Cropped spectrogram images are then grouped into separate folders according to type. The produced datasets are then fed into training using 500,000 training steps. The algorithm is integrated with TensorFlow CPU version 1.5 and Inception V3 model to take advantage of its astonishing way on how it analyzes images. The deep learning neural network involves a bottleneck layer which uses lesser neurons to reduce the number of feature maps in the network to get the best loss during training. In order to have a faster learning rate, the cross-entropy cost function was used. The final accuracy test from the training reached as high as 96.8%. An actual test for identified PAF and healthy datasets from Physionet.org were performed and all are correctly predicted and thus could be able to classify other different diseases based from converted ECG numerical values. Furthermore, this paper established a low-powered workstation’s requirement for implementation because it only requires at least a dual core processor and 2 GB of RAM.

In recent years, the application of Fluorescence Diffusion Optical Tomography (FDOT) technology in pre-clinical experiments is increasing. It is a non-invasive small animal functional image based on diffusion optical tomography (DOT), which uses photons to absorb and scatter the properties of different materials to reconstruct the structure of the small animal sections. Using this technology can reduce the amount of small animal used and accelerate the research courses from academic researches to clinical applications. However, the potential of this technique is currently limited by its poor spatial resolution. In this work, we develop a dual-modality imaging system combining three-dimensional (3D) fluorescence tomography with ultrasound (US) imaging. This design includes an electron-multiplying charge-coupled device (EMCCD), an ultrasound transducer and a fiber-coupled laser on a planar platform. A customized fluorescence/US system was used to reconstruct the 3D fluorescence tomography from optical surface images in position of the inclusions from US signals by using Near-Infrared Fluorescence and Spectral Tomography (Nirfast). Ultrasound B-mode imaging was used to obtain the structural information to precisely extract the tissue boundary of a sample and improve fluorescence reconstruction. The advantages of using this system are noninvasive, easy-to-use and good contrast to soft tissue. We validated the system with meat and a 4T1 tumor nude mice. From the FDOT image and the line profile, it can be seen that the edge information added to the ultrasound is sharper, and the soft tissue and the tumor can also get good results at the same time. The reconstruction results show the combined fluorescence/US system can effectively localize the tumor and drug metabolism. It is very helpful for the study of tumor location and the development of cancer drugs in the small animal study.

Cancer is the second leading cause of death worldwide. In order to reduce this burden, new strategies need to be implemented. With the evolution of computational techniques, the amount of clinical data available has increased considerably. However, these data are not always easily accessible, with some barriers of different nature preventing their retrieval and analysis. With the aim of better understanding these barriers and eventually support decision making through digital tools, a literature search has been conducted and, according to the analysis of the resulting bibliography, a survey to validate the literature findings has been distributed to clinical and computer science experts who work on cancer research in different countries. The answers received allow us to identify the main issues that need to be addressed, which are analyzed and presented in this paper. This work is carried out in the context of BD2Decide European Research project.

Peripheral arterial occlusive disease has a high risk to occur in the lower extremity which causes low oxygen saturation level in muscle tissue, especially after exercise. A muscle tissue oxygen saturation measurement system was developed to detect the insufficient blood supply to lower limb.The diffusive reflectance travelled though deep tissue is measured at two separated distance from the light sources. The individual differences in skin tissue would be cancelled out between these two measurements. Three near–infrared wavelengths are used to measure and calculate the oxygen saturation in deep muscle tissue. The stability of in vivo measurement is tested at different body postures in this study. The detection of the change in oxygen saturation was also tested by an artery occlusion experiment at the lower extremity. The results match the general physiological condition and the reliability of the system is confirmed.

In recent years, the prevalence of peripheral arterial disease (PAD) has increased. This disease is related to arterial occlusion in the heart and brain. Many researches mentioned, measurement of muscle tissue oxygen saturation by near-infrared spectroscopy (NIRS) has a different phenomenon for normal people and patients. In order to quantify the muscle oxygen saturation measured by near-infrared spectroscopy. In this research, we use a in-house made multi-wavelength and two light-detector distances to measure the liquid phantom with adjustable oxygen saturation. LED polarizer wavelength is 740 nm, 808 nm and 850 nm. Two light-detector distances are 32 mm and 40 mm. Liquid phantom using purified red blood cell from pig blood and intralipid to mix the blood cell solution. The liquid phantom simulated lower limbs muscles optical properties. Use yeast and oxygen to change oxygen saturation in the liquid phantom and continuous measurement at the same time. Measurement result show, absorbance at 740 nm and 850 nm will change with oxygen saturation, but absorbance at 808 nm will not. In the other side, between two light-detector distances the trend of change is small. This result is consistent with the absorption spectrum of hemoglobin to calculate a calibration curve for muscle oxygen saturation quantification in a in-house made machine. In the future, it will be embedded in the signal of clinical measurement to help PAD patient’s classification. This classification will become one of the indicators for doctor diagnosis.

Users play games because of fun and relaxation. But some users engage in the online game world, and they give up their educational or work opportunity and neglect their duties. These persons have the symptoms of Internet gaming disorder (IGD). Individual with IGD plays different online game scenes with dynamically psychophysiological control for a long period of time. However, the empirical studies focus on the long period of playing experiences or the short period of physiological responses for investigating the psychophysiological properties of IGD. Few studies discuss the instantaneous psychophysiological responses of persons with IGD. Moreover, individual differences in reaction time may influence the time-variate properties of IGD. It makes investigation of instantaneous psychophysiological responses more difficult. On the basis of the concept of multi-modal pressure-flow method, we propose an index of instantaneous phase delay (IPD) as a modulation of time for observing the instantaneous coordination of respiratory wall movement during watching game video. 19 and 21 persons with high-risk IGD (HIGD) and low-risk IGD (LIGD) were participant in this study, respectively. Preliminary result shows the negative correlation between the IGD questionnaires and IPD (within 3 cycles). Our finding indicated that individuals with HIGD may rapidly modulate psychophysiological response during negative game stimuli compared with individuals with LIGD. We suggested that IPD may be an potential index to assess IGD within 3 cycles. The findings should make an important contribution to advance the understanding of instantaneous regulation mechanism for IGD. Recruiting more participles is needed to verify this finding in the near future.

Polysomnography is the main diagnostic instrument for sleep apnea, but it is not convenient for patient to take frequent measurements at home. An in-home sleep test is a solution to help physicians to collect more information. Sleep medicine societies suggest that a home sleep-apnea testing device should measure at least oxygen saturation and breathing signals. To reduce the discomfort of measurement during sleep, the device should be also as small as possible. One possible measuring site for both signals is at the nose. The accuracy of oxygen saturation measurement relies on the quality of photoplethysmographic signals (PPG). PPG signals of red and near-infrared light measured at the nasal septum and nasal wing were compared. PPG signals were also measured at different body postures includes standing, sitting and supine.Although nasal septum and wing are both thin layer of tissue, the perfusion index is greater and has a larger variation at nasal septum than at nasal wing. Although the PPG at nasal wing is relatively stable and not greatly affected by the autonomic nervous regulation, the change in signals with oxygen saturation is also small. Therefore, the nasal septum is a better position for measuring oxygen saturation at the nose.

Electrohysterography (EHG) has been recently applied as one of diagnostic tools for pregnant women. In this study, six time-domain features commonly applied to EMG data, namely, the root mean square, the mean absolute value, the ν-order, the average amplitude change, the difference absolute standard deviation value and the zero crossing, are applied to EHG data recorded from pregnant women who delivered at term or prematurely. The correlation between the time-domain features of EHG data and their corresponding time of recordings are examined using Pearson’s correlation coefficients. From the computational results, it is shown that the mean absolute value, the ν-order, the zero crossing, and the difference absolute standard deviation value are the time-domain features of EHG data that exhibit the strongest linear correlation with the time of recordings for each class of EHG data. Furthermore, the characteristics of time-domain features of EHG data associated with preterm and term births are shown to be different according to their correlation with the time of recordings.

Dysarthria is a communication disorder common in people with damaged neuro-muscular apparatus resulting from events such as stroke. For a dysarthric speaker, voice conversion (VC) is one of the well-known approaches to improve speech intelligibility for a dysarthric speaker. Most of the well-known VC methods focus on converting amplitude features without phase information. Previous studies indicated that phase is an important factor in the speech signal. Therefore, we are interested in adding the correct phase information to VC for dysarthria speech. The results of automatic speech recognition and spectrum analysis show that intelligibility is improved by replacing the dysarthria phase with the normal phase during the synthesis step. It implies that the correct phase information must be considered for the dysarthria VC system.

In recent therapeutic studies, light therapy has been used to treat seasonal depression disorder in countries where there is insufficient daylight during winter. Previous light therapy studies have used one treatment for all patients, irrespective of individual differences and drug control. Although light therapy has been extended to uses in a few psychiatric treatment programs for major depressive disorder (MDD), there is a lack of consistent research and conclusion regarding its effects of different combinations of lights and the neural mechanism underlying the improvement after therapy. The present study intends to propose several combinations of lights using the beneficial physical properties in prior research and validate the efficacy of the therapies with neurophysiological techniques. Twelve patients suffering from major depressive disorder were enrolled in the study. Five were in the experimental group who will receive the two-month light therapy, with 1 female and 4 males, aged from 38 to 63 years old (mean = 49, SD = 8.51). Seven were in the control group, with 5 females and 2 males, aged from 32 to 53 years old (mean = 42.71, SD = 8.56). All participants were scanned when they were enrolled in the program, a month after pure drug treatment. The control group were scanned a month after their light therapy, and the last time after the light therapy were completed. Results revealed that the default mode network and the salience network were altered after the therapy. The self-report of life quality was better after the therapy. The conclusion is that light therapy could have a lasting effect on the brain by changing the neural connectivity, which led to the improvement in patients with MDD.

A preliminary design of integrated RFID aperture and washing chamber shield for real-time cleaning monitoring in healthcare laundry system is proposed. The installation of RF based monitoring system includes sewing attachment of the RFID tags to the healthcare clothes and the setting up of RFID shield, aperture, and reader to the washing tub, and open area of the washing machine. During the wash, the clothes circulation was tracked by filtered RF transmission between the tags and reader. Software was designed to evaluate the circulation e.g. rotating, sinking, and floating of the various clothes in the batch and update the circulation level in real-time. Two types of conventional healthcare washing machines i.e. agitator and pulse flow were selected for the experiments. The Received Signal Strength Indicator (RSSI) was evaluated on each combination of RFID design, washing condition and clothes circulation level. The cleaning performance of each combination of washing condition and circulation was evaluated by using Microbiological (RODAC plate count) testing. Design of experiments (DOE) methodology of 25 was used to examine the relationship between washing circulation and cleaning performance on representative healthcare laundry machines. 5 trials were repeated at each experimental condition before the repeatability of RSSI was examined. The design providing best repetitive RSSI is presented.

This research paper proposed the measurement and calculation of upper-limb joint power transmission efficiency derived from propulsion efficiency and upper limb joint power efficiency during manual wheelchair propulsion. Portable measurement system was used to collect propulsion force at the push rim and inertia of each upper-limb segment (dominant limb). The relationship between upper-limb joint power transmission efficiency and the overuse of upper limbs was established on the manual wheelchair propulsion experiments in experienced, inexperienced and joint problem recovery users riding on level and slope paths. The data collected by using portable instruments were validated by that collected by using motion analysis system. The diagnosis of upper-limb overuse signifies the highest risks in inexperienced wheelchair riding group.

A verification of individual margins of instantaneous dynamic stability in elderly volunteers is presented. The modified condition and margins of instantaneous dynamic stability permitting the estimation in elderly are proposed. By utilizing in experimental protocols of conventional tests of mobility, i.e. Up-and-Go activity, and balance, i.e. mini-BEST, the periodic stability indices were estimated in elderly without introducing fall prone incidence. The margins of instantaneous dynamic stability or the risk of falls of each elderly volunteer was compared to the healthy young group. The accuracy of fall risk estimation by using the current method tested in fall incidences of young and healthy volunteers was presented.

Visible light communication (VLC) can play a versatile role in future IoT (Internet of Things) - based communication services. In this paper, we first describe the latest encryption technology required to guarantee security specific to VLC and IoT to prevent eavesdropping and then the cyber-physical VLC applications of this technology.This paper clarifies the latest communication concepts that can be implemented with the current technological level of color shift keying (CSK) communications. The CSK receiver and transmitter can identify a specific SDM frame shape, chromaticity coordinate assignment area in a color diagram, and the CSK code. After identifying both the chromaticity coordinates of CSK cells and the frame shape, the corresponding CSK cell content is paraphrased into the original data by different mapping tables. To introduce CSK communications for commercial purposes, security must be enhanced by adopting encryption technologies. It should also be noted that changing the shape of the CSK frame can significantly improve security and the error-free performance can be attained by increasing the number of CSK symbols in a frame.We also describe new communication service concepts that can be implemented with the current technological level of CSK communication. We propose smart glasses applications using cyber-physical space by combining the VLC technology and a computer database, leading AR applications.

The dynamic roles of photosensitizer (PS) concentration and light intensity were measured and analyzed. The efficacy of photodynamic therapy (PDT) and cell viability (CV) are measured (in vitro) and analyzed by analytic and numerical modeling. For a fixed PS concentration, CV is a nonlinear deceasing function of light intensity and exposure time; for a fixed light intensity, higher PS concentration achieves higher efficacy or smaller CV (at steady-state), in consistent to our analytic formulas. Finally, anti-cancer efficacy may be enhanced by the resupply of PS and/or external oxygen.

Circadian rhythms are physiological and behavioural processes that typically recur over 24-h periods.Researchers show that circadian disruption, a marked break in normal 24-h cycles of circadian rhythms, can cause serious health problems. It could lead to critical illness, cancer, stress, myocardial infarction, diabetes, hypertension and arrhythmias.Today, circadian rhythms are monitored using blood, salivary and urine hormone tests, such tests are not practical at home and do not provide continuous real-time monitoring. Combining signal processing and artificial intelligence with commercial sensors embedded in smartwatches or clothes that measure physiological and behavioral attributes offers unprecedented and as yet unexplored opportunities to monitor circadian rhythms in real time. This paper presents the initial steps towards the development of a model for real-time monitoring of the circadian rhythms. This model will contribute to transform medicine from primarily intervention-focused to predictive and preventative. Preliminary analysis shows promising results to automatically classify cortisol levels as high or low, based on behavioral and physiological signals monitored by non-invasive wearable sensors.

Microvascular perfusion on the foot bottom in 39 subjects (Control group: 23 healthy participants; Patient group: 16 patients with diabetes mellitus) was measured with the laser Doppler flowmetry (LDF). Each subject was requested to perform a non-invasive provocation of 37 min, including 8-min baseline, 3-min ankle occlusion, 6-min post-occlusive reactive hyperemia (PORH), and 20-min heating (42 °C) period. By using the wavelet transform, we calculated the power spectral densities (PSD), on one-minute basis, of the 37-min LDF signal. The results indicated that the PSD corresponding to the endothelial NO-independent (PSDENDO1) and NO-dependent (PSDENDO2) metabolic activities varied with time in both Control and Patient groups. Patient group showed less PSDENDO1 and PSDENDO2 than those in Control group. In summary, endothelial dysfunction in peripheral microcirculation exists in diabetes patients, apparently as compared with healthy participants.

The rise of wearable technologies is enabling novel ways of assessing balance and risk of falling in later life. Wearable inertial sensors are a promising addition to clinical balance assessment tools since they provide an objective and accurate fall risk assessment. Moreover, wearable devices also enable the ambulatory monitoring of physiological and behavioural variables, which can be used to infer health status and health-related behaviours linked to impaired balance and fall risk (e.g. sleep disturbances and poor sleep quality). This situation could potentially expand the prevailing paradigm in fall prevention, from the current one mainly involving the occasional assessment of risk factors to a new paradigm also including the continuous monitoring and detection of short-lived factors that might result in an imminent fall. Additionally, the diffusion of the dynamical systems theory and methods within the medical research community is inspiring a new approach to the study of ageing and balance in older adults. In particular, nonlinear signal analysis methods could potentially provide with further information on the underlying control mechanisms in ageing and produce more sensitive measures of fall risk. However, there are several challenges in the adoption of these devices and methods, which still preclude a firm conclusion on their clinical value. This paper summarises three studies performed to address some of these challenges and distils the lessons learnt from them. Collectively, the findings of this research confirm that these sensors and methods could improve currents tools and practices for balance and fall risk assessment, and provides some insights concerning their optimal use.

Electrocardiogram (ECG) is a graph that depicts blood circulation through the heart. ECG is also used for depicting the state of health of an individual and is helpful in disease diagnosis. The target of this work is to check the application of curve fitting on ECG signals based on the Fourier series analysis method. When ECG signals are approximated by the Fourier series model, the fitting for the cardiac cycle is used for judging arrhythmias. The data used here was sourced from the MIT-BIH arrhythmia database, and only ECG recordings were utilized for the purpose of this study. The study has presented efficient methods for signal identification with the help of fitting parameters and ECG classification.

Breast calcification or microtumors screening can early detect breast cancer that can make the disease easier to treat. At present, the segmentation of breast calcifications relies on the delineate by doctors. The process is time-consuming, and the benefits are not readily apparent. None of the paper has been discussed on combining automatically delineate and classify the breast calcifications to benign or malignant in previous research. According to the above reasons, we proposed an approach on combining Cascade Adaboost with CNN to delineate breast calcifications in mammogram and classify breast calcifications to benign or malignant by the CNN we trained. The ability of classification in Cascade Adaboost algorithm is better than Adaboost algorithm, it can significantly reduce the time cost by classification in CNN and speed up the process time. In this paper, we compare our method with the architecture of R-CNN combining CNN, and the experimental results show that by using Cascade Adaboost combined with CNN can detect calcification more accurately and classify it into benign or malignant. We hope that by using the approach in this work can help doctors to detect and diagnose breast calcifications in less time.

Taking advantage of high energy and spatial resolution and high count sensitivity, the ultrafast cardiac γ-camera with cadmium-zinc-telluride (CZT)-based detectors has become popular in practice for myocardial perfusion imaging (MPI). The shorter imaging time with better imaging quality using CZT detectors compared to conventional ones makes it feasible to perform the left ventricular ejection fraction (LVEF) in the meantime of doing 201Tl MPI. The aim of this study was to compare LVEF from 201Tl MPI using a CZT with that from first-pass radionuclide angiography (FPRA) using a NaI camera. A total of 117 patients (aged 81 ± 13 years old) were collected. All underwent 201Tl MPI using a CZT camera (Discovery NM530c) and FPRA using a conventional camera (Symbia E Signal Head System) in 2 wks. Correlations of LVEF obtained from these two examinations were evaluated by SPSS 20.0 statistical software.Our results showed that the mean LVEF measured from MPI and FPRA were 54 ± 18% and 51 ± 16%, respectively. A good linear correlation was found between both methods (r: 0.861, p < 0.0001). It also showed a good agreement and LVEF prediction rates (k = 0.83) obtained from these two measurements. Tl-201 MPI with CZT camera is thus capable to offer a reliable clinical LVEF references.

The use of accelerometer (ACC) sensors above the carotid artery provides an interesting approach to pulse detection during Cardiopulmonary Resuscitation (CPR) efforts. In order to study the basic feasibility of these ACC sensors in a resuscitation scenario, a protocol was designed with the aim of simulating characteristics present in a real-life scenario under controlled conditions. Using this protocol, a dataset of 12 healthy volunteers’ signals was created. For each subject two ACC signals, electrocardiogram (ECG) and photoplethysmography (PPG) were measured synchronously. Additionally, a dataset from a previous study of 5 patients undergoing real-life CPR was available allowing for a comparison between the behavior of the simulated acquired data with real-life signals. Using these two datasets, technical solutions were developed with two different classifiers discriminating artefacts, compressions, pulse and absence of pulse.

Congestive Heart Failure (CHF) is a serious chronic cardiac condition that brings high risk of urgent hospitalization and could lead to death. In this work we show how all the input clinical parameters for classifying CHF using Machine Learning can be acquired. The requested input are Blood Pressure, Heart Rate, Brain Natriuretic Peptide, Electrocardiogram, Blood Oxygen Saturation, Height, Weight and Ejection Fraction. The next step will be designing a novel device and connecting it to our Machine Learning classifier. A particular attention will be put to the assessment of electromagnetic compatibility (EMC) with other devices, taking into account that this new device will be used in many different settings (home, outdoor, etc.).

Human body communication (HBC) has the advantages of low power consumption, low radiation and anti-interference ability, which has broad application prospects in entertainment and health care. Accurate human channel characteristics will contribute to the development of HBC. This paper explores the effect of channel length and electrode size on human body channel phase characteristic. A galvanic coupling human body communication experimental platform was built. The measurement results show that in the low frequency band, the phase transition is less than the high frequency. When the frequency is in the range of 200 kHz to 300 kHz, the phase will oscillate. Increasing the electrode size can improve phase oscillation. This paper provides a reference for the application of human body communication in long channel and the miniaturization design of transceiver.

Continuous Positive Airway Pressure (CPAP) machine is a form of positive airway pressure ventilator, which utilizes mild air pressure on a continuous basis to keep the airways continuously open in people who are not able to breathe spontaneously on their own. The CPAP machine is widely used for sleep apnea patients. This paper presents the development of active noise control (ANC) system for reducing the noise from CPAP machine. By integrating loudspeaker and microphones, we develop feedback ANC structure and filtered-X least mean square (FXLMS) algorithm using the Texas Instrument (TI) TMS320C6713 starter kit. Real-time experimental results show that the proposed method reduces the noise of CPAP machine and achieves global cancellation of the noise.

Dysphonia is a vocal impediment that appears as a symptom of Parkinson’s disease, and can be used for its diagnosis. Among the important measurements for dysphonia detection are jitter, shimmer, fundamental frequency (F0), Harmonics to noise ratio (HNR) and noise to harmonics ratio (NHR). The frequency space of the speech signal is used to detect these five dysphonia measurements, through this space the acoustic markers jitter, shimmer and F0 are calculated. In this article, an evaluation of the detection of acoustic markers is presented through the mathematical methods of the Constant Q Transform (CQT) and the Mel Frequencies Cepstral Coefficients (MFCC) in speech signals of patients with Parkinson’s disease. The classifier method Support Vector Machine (SVM) is used to detect the Biomarkers. According to the results, the CQT method and MFCC method (57% and 62% precision respectively) which is a promising results for Parkinson’s disease diagnosis by the detection of Dysphonia measurements.

There are a number of methods and diagnostic tools for assessing the health of the heart. The systolic time intervals (STI) have been clinically useful parameters representing cardiac cycle and measuring the ventricular performance. In this study, a continuous wavelet transform (CWT) based approach for quantifying the systolic time intervals, in particular, the pre-ejection period (PEP) and the left ventricular ejection time (LVET), using electrocardiogram (ECG) and phonocardiogram (PCG) signals are proposed. The proposed CWT-based STI quantification approach is composed of three main stages: ECG signal processing, PCG signal processing, and computation of the systolic time intervals. The proposed CWT-based STI quantification approach is validated using ECG and PCG data recorded from both healthy subjects and subjects suffering from various cardiovascular diseases. The computational results suggest that the proposed CWT-based STI quantification approach has a considerable capability for clinical applications. The means of average absolute errors on PEP and LVET quantifications are, respectively, 13.4251 ms and 27.1348 ms. The best total score achieved is 868.9192.

The systolic time intervals of hearts are related to health. People with myocardial dysfunction will have a longer pre-ejection period (PEP) and a shorter left ventricle ejection time (LVET) than healthy people. The purpose of this paper is intended to detect PEP and LVET accurately from electrocardiography (ECG) and phonocardiogram (PCG). Generally, there are several kinds of noises from environment or breathing in PCG. It is necessary but difficult to extract the best signals we want. Our approach is to use a simple DSP-based method in PCG to detect aortic valve opening (AVO) and aortic valve closure (AVC) times as well as R-peak time in ECG. Then, the PEP and LVET can be calculated. We evaluated PEP and LVET of 72 files from 46 people. To the annotated data, the PEP range is around 5 ms to 100 ms; the LVET range is around 170 ms to 380 ms. Our PEP results have 63.96% accuracy within 20 ms absolute error and 91.74% accuracy within 40 ms absolute error; LVET have 75.48% accuracy within 40 ms absolute error and 93.53% accuracy within 80 ms absolute error.

To determine the pre-ejection period (PEP) and the left ventricular ejection time (LVET) thorough heart sounds and ECG are major tasks in this paper. The first step was to determine the event time of PEP, which detected the feature point of the first heart sound (S1) of PCG about 0.02–0.07 s after the R-peak of the ECG, and then obtained the prominent peaks of PCG during changes of signal slope with drastic change of peak value. The second step was taking R-peak to define a period of sound signal, making LPC and FFT, and moving certain few points to make another section, which was repeated to find out changes from the coefficient and frequency. Taking the sections with changes occurred, and FFT results as references got the time of PEP and LVET. Comparing the proposed results to the annotations, the average error of PEP detection is approximately 36.4 ms, and that of LVET is approximately 6.95 ms. With the error of the PEP, 83.4% and 33.9% accuracy are achieved within the time of 40 ms and 20 ms. With the error of the LVET, 94.2% and 25.8% accuracy are achieved at the time of 80 ms and 60 ms. The advantage of this method is that although the signals are very small, finding out the peak value of PCG, and analyzing the PEP and LVET are not hard tasks. Another advantage is that if the noise of the signal is not big enough to affect the original characteristic, the ECG signal could be applied to find out the peak value perfectly, and could be cut into piece by piece. The “NaN” point is hard to be defined in the testing data. Some issues are required to be improved or verified by other methods.

Non-contact heart rate measurement has been widely utilized in multiple applications. The Eulerian Video Magnification algorithm proposed by MIT CSAIL group in 2012 can be used to magnify the subtle color variation and small motion in videos and can be used to extract cardio-features through photoplethysmography method. In this study, we intend to improve the accuracy of heart rate prediction for the Eulerian Video Magnification method. With the selected region of interest and the peak detection algorithm, we found out that the signal of the Y component in YIQ color spectrum is more consistent than that of the I component in terms of heart rate estimation. The result also demonstrated that the heart rate extracted under 30 frames per second (fps) was more accurate than which extracted under 60 fps. With an illumination level higher than 1500 lx and a frame rate of 30 fps, the error of heart rate extraction compared to oximeter measurement was 5% while using GoPro Hero 6 for recording. Further data processing and false peak detection are necessary for accurate heart rate variability characterization.

Intravascular Optical Coherence Tomography (IVOCT) is a modality which gives in vivo insight of coronaries’ artery morphology. Thus, it helps diagnosis and prevention of atherosclerosis. About 100–300 cross-sectional OCT images are obtained for each artery. Therefore, it is important to facilitate and objectify the process of detecting regions of interest, which otherwise demand a lot of time and effort from medical experts. We propose a processing pipeline to automatically detect parts of the arterial wall which are not normal and possibly consist of plaque. The first step of the processing is transforming OCT images to polar coordinates and to detect the arterial wall. After binarization of the image and removal of the catheter, the arterial wall is detected in each axial line from the first white pixel to a depth of 80 pixels which is equal to 1.5 mm. Then, the arterial wall is split to orthogonal patches which undergo OCT-specific transformations and are labelled as plaque (4 distinct kinds: fibrous, calcified, lipid and mixed) or normal tissue. OCT-specific transformations include enhancing the more reflective parts of the image and rendering patches independent of the arterial wall curvature. The patches are input to AlexNet which is fine-tuned to learn to classify them. Fine-tuning is performed by retraining an already trained AlexNet with a learning rate which is 20 times larger for the last 3 fully-connected layers than for the initial 5 convolutional layers. 114 cross-sectional images were randomly selected to fine-tune AlexNet while 6 were selected to validate the results. Training accuracy was 100% while validation accuracy was 86%. Drop in validation accuracy rate is attributed mainly to false negatives which concern only calcified plaque. Thus, there is potential in this method especially in detecting the 3 other classes of plaque.

Diabetic peripheral neuropathy DPN is the most frequent complication with people with diabetes and affects approximately half of this population. This is reflected in the reduction of the transmission of vibratory sensitivity, proprioceptive and of reflexes osteotendinous, affecting sensory and motor skills. People with DPN are up to 23 times more likely to have falls and 15 times more likely to report an injury compared to healthy older adults. Falls have significant consequences, such as temporary or permanent physical disability, which decreases the quality of life of the elderly and represents an increase in mortality. Therefore, a pilot study is proposed to assess the risk of falls in older adults with DPN by implementing a predictive system based on the clinical information of the AGS/BGS guidelines, the oscillation of the pressure center (CoP) at rest and the Timed up and Go test (TUG) with the use of inertial sensors. A sample of convenience of 19 participants (13 control group and 6 with DPN) older than 45 years, were cited to perform data acquisition, which were used as inputs to evaluate four fall risk classification techniques: K-means, support vector machines, K nearest neighbors and neural networks, obtaining a precision = 90.9%, sensitivity = 80% and specificity = 100%. Therefore, this study suggests that it is possible to evaluate the risk of falls in older adults with DPN through the clinical information of the AGS/BGS guidelines, the oscillation of the CoP at rest and the TUG test with the use of inertial sensors and it also has the potential to be implemented in future studies with larger populations.

Peripheral Diabetic Neuropathy (PDN) produces nerve damage in lower limps and foot, and therefore, a decrease of information by the proprioceptive system to maintain postural stability denoting greater body sway. Due to postural stability pilot study was develop by using the anteroposterior center of pressure in 60 participants from Bogota, Colombia: 30 diabetics (19 female and 11 male) with PDN and 30 healthy controls (19 female and 11 male) matched by gender, age, height and weight. Wilcoxon test (p < 0.05) was used to find significant differences in linear parameters such as: excursion, velocity, Root mean square RMS, maximum and minimum amplitude. Plantar sensitivity was evaluated with the monofilament test of Semmes-Weinstein 5.0. All Diabetic Peripheral Patient DPN participants showed higher values in all parameters with maximum in excursion and RMS values and all DPN presented at least one insensitive zone at the plantar level. In static position DPN participants showed higher body sway and tends to stay on the anterior axis to maintain postural stability. This pilot study stablished what other previous literature have report that DPN disease directly impacts in postural stability becoming an important risk factor for falls.

Falls can have a major impact on physical and psychological health on elderly people who experience them. To reduce the negative consequences of a fall event, automatic fall detection systems are being developed to correctly identify when a person falls and alert the caregivers to provide assistance on time. Performance of fall detection algorithms is tested with datasets containing measurements of falls and regular activities of daily living. In this work we acquired fall signals with accelerometer sensors and compared them with digitized fall signal records from 6 different datasets. Additionally, three threshold-based algorithms for fall detection were implemented and their performance was tested with the analyzed datasets. The results suggest that a heterogeneity among the fall data in distinct datasets exist and that it affects the performance measures of tested datasets.

Continuous glucose monitor (CGM) is a wearable body sensor device that automatically measures glucose at regular intervals. CGM records can be used to analyze data after the collection or for predicting certain trends in the records in real-time. Nevertheless, missing values are common in CGM records. In this paper we discuss possible methods to replace the missing values in short-length gaps, in both real-time and after the collection use.

Delineating organs for a long time may cause exhaustion to radiologist’s eyes and mental health, it could lead to results that show different sizes of organs with therapeutic target volume. In this work, we expand on the idea of automatically delineating the organs in Computed Tomography (CT) images of head and neck through the generative adversarial network, which is a deep learning algorithm. In image preprocessing, we generate a bitmap (BMP) image by the combination of CT image and RT structure (RS) file and input it to generator network, which will improve the color and texture quality, last generate a fake Radiation Therapy (RT) image. Finally, the discriminator network takes the fake RT image as an example to compare with the original RT image. To build the predictive model, we continuously train this model to let it learn the rules of delineating organs in CT image, generating more and more images that are similar to the original samples. The approach that proposed in this paper is actually well applied in medicine, and the results of testing are similar to the selected organs or therapeutic targets’ volume that was delineated by the radiologist. We can see that it not only effectively reduces the false positive rate but also promises in applying to other related images.