Medical Imaging and Informatics

Feature selection and parameter settings for classifiers are both important issues in computer-assisted medical diagnosis. In the present paper, we highlight some of the complexity problems posed by both tasks. For the feature selection problem we propose a search-based procedure with a proven time bound for the convergence to optimum solutions. Interestingly, the time bound differs from fixed-parameter tractable algorithms by an instance-specific factor only. The stochastic search method has been utilized in the context of micro array data classification. For the classification of medical images we propose a generic upper bound for the size of classifiers that basically depends on the number of training samples only. The evaluation on a number of benchmark problems produced a close correspondence to the size of classifiers with best generalization results reported in the literature.

The Higher Education in Clinical Engineering (HECE) of the University of Trieste constituted in 2005 the Open Three Consortium (O3), an innovative open-source project dealing with the multi-centric integration of hospitals, RHIOs (Regional health information organizations) and citizens (care at home and on the move, and ambient assisted living), based on about 60 HECE bilateral cooperation Agreements with Hospitals, Medical Research Centers, Healthcare Enterprises, Industrial Enterprises and Governmental Agencies and on the International Networks ABIC-BME (Adriatic Balcanic Ionian Cooperation on Biomedical Engineering) and ALADIN (Alpe Adria Initiative Universities’ Network). The collaboration with multiple open-source solutions has been extended, starting an international cooperation with the open-source based company Sequence Managers Software, Raleigh, NC, United States. The O3 Consortium proposes e-inclusive citizen-centric solutions to cover the above reported three main aspects of the future of e-health in Europe with open-source strategies joined to full-service maintenance and management models. The Users’ and Developers’ O3 Consortium Communities are based mainly on the HECE agreements.

Emerging technologies are transforming the workflows in healthcare enterprises. Today, several vendors offer holistic web-based solutions for radiologists, radiographers and clinicians - a single platform for all users. Besides traditional web, streaming technology is also emerging to the radiological practice in order for improving security and enabling the use of low network bandwidths.

The technology does not set limitations any more: today, the digital workplace knows no boundaries; remote reporting, off-hour coverage, virtual radiologists are all ways to offer imaging services in a non-traditional way. The challenge, however, is to provide trust over distance - across organizational or even national boundaries. In the following three different aspects important in building trust in remote reporting are discussed: 1) organizational change issues, 2) continuous feedback and 3) legal implications.

Stereotactic operations were performed with the frameless stereotactic instrument (named as CAS-R-2) manufactured by ourselves rather than traditional stereotactic frame. The aim of this study was to assess the clinical usefulness, accuracy and safety of the frameless stereotactic instrument. We retrospectively reviewed 2011 patients aged between 0.2 to 89 years (with mean of 30.7 years) with CT/MRI image-guided frameless stereotactic surgery between January 1997 to April 2007. The accuracy of position and improvement of symptom was observed. The surgical procedures were successful. All targets were pointed accurately in just one go during the operation. Follow-up being performed 3 to 48 months (averaged 24 months) after the operation, the total effective rate was 93.3% without serious surgery-related complications. Compared with the traditional frame stereotactic operations, this method has some advantages, such as releasing the patient’s pain, convenient to the doctors, extending the range of indications and increasing the safety and effectiveness of the operations.

A level set method based on the Bayesian risk is proposed for medical image segmentation. At first, the image segmentation is formulated as a classification of pixels. Then the Bayesian risk is formed by false-positive and false-negative fractions in a hypothesis test. Through minimizing the average risk of decision in favor of the hypotheses, the level set evolution functional is deduced for finding the boundaries of targets. To prevent the propagating curves from generating excessively irregular shapes and lots of small regions, curvature and gradient of edges in the image are integrated into the functional. Finally, the Euler-Lagrange formula is used to find the iterative level set equation from the derived functional. Comparing with other level-set methods, the proposed approach relies on the optimum decision of pixel classification; thus the approach has more reliability in theory and practice. Experiments show that the proposed approach can accurately extract the complicated shape of targets and is robust for various types of images including high-noisy and low-contrast images,

CT

,

MRI

, and ultrasound images; moreover, the algorithm is extendable for multiphase segmentation.

An intelligent deformable model called worm model is constructed. The worm has a central nervous system, vision, perception and motor systems. It is able to memorize, recognize objects and control the motion of its body. The new model overcomes the defects of existing methods since it is able to process the segmentation of the image intelligently using more information available rather than using pixels and gradients only. The experimental results of segmentation of the corpus callosum from MRI brain images show that the proposed worm model is able to segment medical images automatically and accurately. For those images that are more complex or with fragmentary boundaries, the predominance of the worm model is especially clear.

A beam-hardening effect is a common problem affecting the quantitative ability of X-ray computed tomography. We develop a statistical reconstruction for a poly-energetic model, which can effectively reduce beam-hardening effects. A phantom test is used to evaluate our approach in comparison with traditional correction methods. Unlike previous methods, our algorithm utilizes multiple energy-corresponding blank scans to estimate attenuation map for a particular energy spectrum. Therefore, our algorithm has an energy-selective reconstruction. In addition to the benefits of other iterative reconstructions, our algorithm has the advantage in no requirement for prior knowledge about object material, energy spectrum of source and energy sensitivity of the detector. The results showed an improvement in the coefficient of variation, uniformity and signal-to-noise ratio demonstrating better beam hardening correction in our approach.

This paper presents an image super-resolution method that enhances spatial resolution of MRI images in the slice-select direction. The algorithm employs Tikhonov regularization, using a standard model of imaging process and reformulating the reconstruction as a regularized minimization task. Our experimental result shows improvements in both signal-to-noise ratio and visual quality.

In this paper, a simple enhancement algorithm for MR head images has been presented. The algorithm is based on histogram equalization but new adaptive reassigning rules have been involved, which approaches a non-linear gray level mapping. Comparing with other existing enhancement algorithms based on equalization, the new algorithm needs not calculate local histograms window by window but dynamically assigning new gray levels according to statistical info in related histogram, which makes the new algorithm natively faster. Testing results on different MR Head images have been reported and compared with several existing algorithms, which have shown that the new algorithm is not only faster but also reached better enhancement results.

Segmentation is one of the most difficult tasks in digital image processing. This paper presents a novel segmentation algorithm, which uses a biologically inspired paradigm known as artificial ant colonies. Considering the features of artificial ant colonies, we present an extended model applied in image segmentation. Each ant in our model is endowed with the ability of memorizing a reference object, which will be refreshed when a new target is found. A fuzzy connectedness measure is adopted to evaluate the similarity between the target and the reference object. The behavior of one ant is affected by the neighboring ants and the cooperation between ants is performed by exchanging information through pheromone updating. The simulated results show the efficiency of the new algorithm, which is able to preserve the detail of the object and is insensitive to noise.

Kernel Principal Component Analysis (KPCA) is one of the methods available for analyzing ultrasound medical images of liver cancer. First the original ultrasound images need airspace filtering, frequency filtering and morphologic operation to form the characteristic images and these characteristic images are fused into a new characteristic matrix. Then analyzing the matrix by using KPCA and the principle components (in general, they are not unique) are found in order to that the most general characteristics of the original image can be preserved accurately. Finally the eigenvector projection matrix of the original image which is composed of the principle components can reflect the most essential characteristics of the original images. The simulation experiments were made and effective results were acquired. Compared with the experiments of wavelets, the experiment of KPCA showed that KPCA is more effective than wavelets especially in the application of ultrasound medical images.

In this paper, we propose a novel scheme for cell nucleus segmentation which is multi-scale space level set method. Under this scheme, all nuclei of interest in a microscopic image can be segmented simultaneously. The procedure includes three stages. Firstly, the mathematical morphology method is used to search seed points to localize interested nuclei. Secondly, based on the distribution of these seed points, a level set function is initialized. Finally, the level set function evolves and eventually stops zero level set contours at the boundaries of nuclei labeled by seed points. The evolution in the last stage is a three phase evolution. In each phase, information of different scale spaces is employed. This method was tested by truthful microscope images of lymphocyte, which proved its robustness and efficiency.

The essential goal for Statistical Shape Model (SSM) is to describe and extract the shape variations from the landmarks cloud. A standard technique for such variation extraction is by using Principal Component Analysis (PCA). However, PCA assumes that variations are linear in Euclidean vector space, which is not true or insufficient on many medical data. Therefore, we developed a new Geodesic Active Shape (GAS) mode by using Principal Geodesic Analysis (PGA) as an alternative of PCA. The new GAS model is combined with Minimum Description Length approach to find correspondence points across datasets automatically. The results are compared between original MDL and our proposed GAS MDL approach by using the measure of Specificity. Our preliminary results showed that our proposed GAS model achieved better scores on both datasets. Therefore, we conclude that our GAS model can capture shape variations reasonably more specifically than the original Active Shape Model (ASM). Further, analysis on the study of facial profiles dataset showed that our GAS model did not encounter the so-called “Pile Up” problem, whereas original MDL did.

This article describes the use of a frequency–based weighting developed for image retrieval to perform automatic annotation of images (medical and non–medical). The techniques applied are based on a simple

tf/idf

(term frequency, inverse document frequency) weighting scheme of GIFT (GNU Image Finding Tool), which is augmented by feature weights extracted from training data. The additional weights represent a measure of discrimination by taking into account the number of occurrences of the features in pairs of images of the same class or in pairs of images from different classes. The approach is fit to the image classification task by pruning parts of the training data. Further investigations were performed showing that weightings lead to significantly worse classification quality in certain feature domains. A classifier using a mixture of

tf/idf

weighted scoring, learned feature weights, and regular Euclidean distance gave best results using only the simple features. Using the aspect–ratio of images as feature improved results significantly.

Health information systems on the web basically support the English language. To access high-quality online health information it is frequently a barrier for non-English speakers or speakers of English as a foreign language. In this work we present a cross-language retrieval system to support Greek users in the medical domain, overcome the language barrier. We have performed a case study on the impact of stemming in the cross lingual retrieval in association with dictionary based query translation techniques. Finally, we conclude with results from a preliminary evaluation of the Greek-English CLIR prototype.

The technology of medical image retrieval in picture archiving and communication systems (PACS) is of great importance. A shape prior algorithm retrieval based on interest point is presented in this paper. Firstly, according to the formulaic composition of a medical image, a Harris point detector is improved to extract some interest points in images. Secondly, by combining invariants for each point and an edge type histogram, the feature vector for matching is constructed. Finally, a strategy for matching vectors is implemented to retrieve medical images. The test results prove the efficiency of this approach.

In neuroscience, grating cells in areas V1 and V2 of the visual cortex of monkeys can respond vigorously to a grating of bars of appropriate orientation, position and periodicity. Computational models of grating cells have been proposed and used to make texture analysis for medical images. To improve the matching precision, the computation models of grating cells were applied to the responses of simple cells and not for the pixel values of the input image. In this paper, the computational models of grating cells is modified to express uncertain information. Multi-valued logic is introduced into the computation of the responses of the grating subunit. Texture pattern is computed by means of the modified computational model of grating cells. Experiments show that the content-based medical image retrieval system using the modified computational model of grating cells has good performance.

A Hospital information system (HIS) has been proposed to respond to some challenges of the complex business process in a hospital. The business processes undergo constant change and the HIS needs to swiftly adapt to reflect these changes. This paper proposes an approach, which is used to define services and solutions for hospital applications according to the business-entity changes in the hospital business processes. The objective of this approach is to minimize the effect of business requirement changes on the system development. Focusing on the transition from the process-oriented to the application-oriented perspectives, some central development considerations are also presented, which can be used to guide the design of service-based interoperability and illustrate these aspects with examples from our current work. This solution will improve the robustness and extensibility of HIS.

More and more healthcare personnel are using computer networks and wireless enabled PDAs (personal digital assistants) in their daily work. This leads to the vision of smart environments with location-based knowledge and information services in healthcare facilities. Location tracking of healthcare workers and patients naturally facilitates the realization of such visions. It is also useful in enhancing the safety of the residents in a nursing facility. This paper introduces the authors’ efforts in applying wireless technology to track the locations of residents in nursing homes. A software client is developed for an industrial location tracking product. The design and implementation of the software client are discussed in this paper. Test results are also reported.

Within the expanding paradigm of medical imaging in Teleradio- logy-Telemedicine, there is increasing demand for transmitting diagnostic medical imagery. These are usually rich in radiological contents, especially in slicing modalities, and the associated file sizes are large which must be compressed with minimal file size to minimize transmission time and robustly coded to withstand required network medium. It has been reinforced through extensive research that the diagnostically important regions of medical images, Regions of Interest, must be compressed by lossless or near lossless algorithm, while on the other hand, the background region be compressed with some loss of information but still recognizable using JPEG2000 standard. Applying on MRI and CT scan images achieved different high compression ratios with varying quantization levels analogously reduced transmission time depending on sources of energy, the MAXSHIFT method proved very effective both objectively and subjectively.

Previous studies have witnessed that complex brain networks have the properties of high global and local efficiency. In this study, we investigated the gender effect on brain functional networks measured using functional magnetic resonance imaging (fMRI). Our experimental results showed that there were no significant difference in global and local efficiency between male and female. However, the gender-related effects on nodal efficiency were found at several brain regions, including the left middle frontal gyrus, right superior temporal gyrus, left middle cingulum gyrus, left hippocampus, right hippocampus, right parahippocampal and left amygdala. These results were compatible with previous findings. To our knowledge, this study provided the first evidence of gender effect on the efficiency of brain functional networks using resting-state fMRI.

The term input function usually refers to the tracer plasma time activity curve (pTAC), which is necessary for quantitative positron emission tomography (PET) studies. The purpose of this study was to acquire the pTAC from the independent component analysis (ICA) estimated whole blood time activity curve (wTAC) using our proposed method: FDG blood-cell-two-compartment model (BCM). We also compared published models, which are linear haematocrit (HCT) correction, nonlinear HCT correction, and two-exponential correction. According to the results, the normalized root mean square error (NRMSE) and error of area under curve (EAUC) of BCM estimated pTAC were the smallest. Compartmental and graphic analyses were used to estimate metabolic rate of FDG (MR

FDG

). The percentage error of MR

FDG

(PE

MRFDG

) estimated from BCM corrected pTAC was also the smallest. The BCM is a better choice to transfer wTAC to pTAC for quantification.

Positron Emission Tomography (PET) is used within neurology to study the underlying biochemical basis of cognitive functioning. Due to the inherent lack of anatomical information its study in conjunction with image retrieval is limited. Content based image retrieval (CBIR) relies on visual features to quantify and classify images with a degree of domain specific saliency. Numerous CBIR systems have been developed semantic retrieval, has however not been performed. This paper gives a detailed account of the framework of visual features and semantic information utilized within a prototype image retrieval system, for PET neurological data. Images from patients diagnosed with different and known forms of Dementia are studied and compared to controls. Image characteristics with medical saliency are isolated in a top down manner, from the needs of the clinician - to the explicit visual content. These features are represented via Gabor wavelets and mean activity levels of specific anatomical regions. Preliminary results demonstrate that these representations are effective in reflecting image characteristics and subject diagnosis; consequently they are efficient indices within a semantic retrieval system.

The serotonin transporters are target-sites for commonly used antidepressants. [

123

I] ADAM is a novel radiotracer that selectively binds the serotonin transporters (SERTs) of the central nervous system. The aim for this study was to evaluate a non-invasive reference tissue model for SERTs quantification using the cerebellum as the indirect input function. The four-parameter model (FPM) was compared with the three-parameter model (TPM) using [

123

I] ADAM dynamic brain SPECT images. The binding potential values derived from both models were the same, but the ratio of delivery (R

1

) in TPM had a smaller standard deviation than the FPM model. In conclusion, the simplified reference tissue model (TPM) was the better choice because of its stability (small standard deviation) and convenient implementation for non-invasive quantification of brain SPECT studies.

Position Emission Tomography (PET) is increasingly applied in the diagnosis and surgery in patients thanks to its ability of showing nearly all types of lesions including tumour and head injury. However, due to its natures of low resolution and different appearances as a result of different tracers, segmentation of lesions presents great challenges. In this study, a simple and robust algorithm is proposed via additive colour mixture approach. Comparison with the other two methods including Bayesian classified and geodesic active contour is also performed, demonstrating the proposed colouring approach has many advantages in terms of speed, robustness, and user intervention. This research has many medical applications including pharmaceutical trials, decision making for drug treatment or surgery and patients follow-up and shows potential to the development of content-based image databases when coming to characterise PET images using lesion features.

Knowledge Based Imaging is suggested as a method to distinguish blood from tissue signal in transthoracial echocardiography. Parametric model for the autocorrelation functions for turbulent blood flow and slowly moving tissue are augmented for in this paper. The model also includes the presence of stationary clutter noise and system white noise. Knowledge Based Imaging utilizes the maximum likelihood function to classify blood and tissue signal. In amplitude imaging blood and tissue are separated by their difference in signal powers. This effect is also present in Knowledge Based Imaging. In addition, this method utilizes the fact that blood flow is turbulent and moves faster than tissue. Some images of Knowledge Based Imaging with different parameter settings are visually compared with Second-Harmonic Imaging, Fundamental Imaging and Bandwidth Imaging [1].

Attenuation (scattering and absorption) of gamma photons in the patient’s body is one of the major limitations among the others in single photon emission computed tomography (SPECT). It reduces quantitative accuracy of measured radioactivity concentration and causes hot rim artifacts in reconstructed images if not corrected for. A variety of approximate attenuation correction methods has been developed or proposed by various groups to date, but all methods have some limitations. In this paper two attenuation correction methods have been investigated and compared. Data are acquired with both the collimators either LEGP or LEHR by scanning R. A. Carlson cylindrical phantom over 360o with an acrylic block of holes simulating hot regions of various sizes at different locations with respect to the walls of the phantom, which was filled with water and Tc-99m solution was uniformly distributed. Results show that the Chang’s attenuation correction method works better as compared to the Sorenson’s method in terms of the linearity in measured counts in hot regions. However, Chang’s method is sensitive with linear attenuation coefficient values and also gives higher standard deviation values particularly in smaller hot regions count density - with LEHR collimator data - compared to the Sorenson’s method.

Median filtering is an important approach in digital image processing for noise elimination or extraction. The time cost and detection quality of the filtering system are two convention measures, depending on the sliding window size. In this paper, an improved median filter, Adaptive Sliding Window – Simultaneous Deleting and Inserting (ASW-SDI) system, is proposed for calcified lesions’ detection in digital mammograms, increasing the quality of detection and also reducing the time cost. It changes the size of sliding windows adaptively and uses the same pixels in two neighboring windows, deleting and inserting a line of pixels in a single array traverse. It is especially appropriate for images with a small quantity of large noises and a mass of salt & pepper noises. In the breast cancer computer-aided diagnosis experiments, ASW-SDI works efficiently in calcified lesion extraction.

A new estimator, Bandwidth Imaging, related to the bandwidth of the ultrasound Doppler signal is proposed as a classification function of blood and tissue signal in transthoracial echocardiography of the left ventricle. An in vivo experiment is presented, where the apparent error rate of Bandwidth Imaging is compared with the apparent error rate of Second-Harmonic Imaging on 15 healthy men. The apparent error rates are calculated from the 16 myocardial wall segments defined in [1]. A hypothesis test of Bandwidth Imaging having lower apparent error rate than Second-Harmonic Imaging is proved for a p-value of 0.94 in 3 segments in end diastole and in 1 segment in end systole. When data was averaged by a structural element of 5 radial, 3 lateral and 4 temporal samples the numbers of segments increased to 9 in end diastole and to 6 in end systole. This experiment indicates that Bandwidth Imaging can supply additional information for automatic border detection routines on endocardium.

Visible Korean Human (VKH) consisting of magnetic resonance, computer tomography, anatomic, and segmented images was created. In the VKH, several techniques were developed and numerous data were acquired. The VKH techniques mainly contributed to the generation of advanced segmented images, Visible Living Human, and Visible Mouse. Also, a software for viewing sectional anatomy, three dimensional images for virtual dissection and virtual endoscopy, was developed based on the VKH data distributed worldwide. The VKH technique and data are expected to promote development of other serially sectioned images and software, which are helpful in medical education and clinical practice.

A great deal of work has been attempted and accomplished based on VCH-F1 (the No.1 Virtual Chinese Human-Female) dataset these years. So far, the anatomic structures of the whole body have been 3D reconstructed and a varieties of further work such as health science education facilities, virtual acupuncture, image-guided neurosurgery, motion simulation etc. have also been developed and preliminarily implemented. In this paper, we will describe the application study of VCH-F1 dataset in our laboratory.

Since visible human visualization using cryosection images is still a challenge for its own difficulties such as color inhomogeneity between adjacent images, most visible human visualizations use pseudo color. In this paper, we propose a method to make cross-section image along human surface homogeneous, and provide a method to reconstruct and visualize 3D visible human with an approximate and reasonable real surface color. The visualization method consists of three components, which are preprocessing (registering image series, obtaining color checkers’ color and removing background), color correction (global color correction with the help of the color checker and local color correction by using adjacent image) and 3D visualization with color (smoothing model and generating color). The experiment on head data of Chinese Visible Human shows that our method is successful for the 3D reconstruction and color visualization.

For establishing a plan of Living Donor Liver Transplantation (LDLT), it is very important to estimate the volume of each liver segment. Usually Couinaud’s classification is used to segment a liver, which is based on the liver anatomy. However, it is not easy to perform this method in a 3D space directly. In this paper, a fast segment method based on the hepatic vessel tree was proposed. This method was composed of four main steps: vasculature segmentation, 3D thinning, vascular tree pruning and classification, and vascular projection and curve fitting. This method was validated by application to a 3D liver from CT data, and it was shown to approximate closely Couinaud’s classification with high speed.

The application of watersnakes algorithm in segmentation of hippocampus MR image has been investigated.This algorithm integrates the watershed transform and the active contour algorithm. The watershed transform, based on mathematical morphology, is powerful and flexible for segmentation. However, it does not allow the characteristics of region boundaries to be included in the way that active contour algorithm does. So, over-segmentation is shown in the result of watershed transform, this phenomenon is even worse for the segmentation of hippocampus. For watersnakes algorithm, the primitive contour of hippocampus can be obtained using watershed transform. Based on energy-driven, the contour of hippocampus can develop into the ultimate result. In the process of energy-driven, the information relating to characteristics of region boundaries is involved.

Recently, the authors introduced least square quantization table (LSQT) method to accelerate the direct Fourier transform to reconstruct magnetic resonance images acquired using a spiral trajectory. In this paper, we will discuss the LSQT further in its adaptability, reusability and choice of the number of groups. The experimental results show that the LSQT method has better adaptability for the different reconstruction cases than the equal phase line (EPL) and Kaiser-Bessel gridding methods. Additionally, it can be reused for reconstructing different images of varied sizes.

Background removal is a critical step in Visible Human Data (VHD) processing, which is the basic of all other researches. In this paper, a new segmentation algorithm based on the hybrid method for VHD background removal has been proposed, which combines a feature based segmentation method with a contour based one. The algorithm first determines the background part and the interested parts of an image at a coarse level by using its colour features, and then obtains a fine segmentation by using a Gradient Vector Flow (GVF) Snake model on the previous initial contour. Our test results on Chinese VHD show that the new algorithm is more robust and accurate than the previous methods.

Finite element analysis is a useful tool for modeling surgical effects on the cornea and developing a better understanding of the biomechanics of the cornea. In this paper we proposed a method of building a physical model of the cornea with finite element method. Firstly, an individual 3D modal model of cornea was constructed. Then the finite element model was built up based on a nonlinearly elastic, isotropic formulation. Finally, the intra-ocular pressure and external pressure were simulated with results of cornea shape changes computed via finite element analysis.

An improved hybrid projection function (IHPF) for precise eye location is presented in this paper. This algorithm combined the advantage of variance projection function (VPF) and hybrid projection function (HPF) by optimizing their weights in the traditional integral projection function (IPF). Two different face databases, BioID face database downloaded from the internet and PSFace database established by our laboratory, were used to test the influence of different projection functions on correctness and relative mean-error of eye locations. The results show that IHPF with optimized proportion factors has a high eye location correctness of 96~100% and the lowest relative error with better face feature location capability.

Light scattering spectra and polarization states can provide important information about skin. To analyse the mechanisms of interaction between skin and light, and the relationship between the changes of light’s characteristics and the variations of skin’s states, a spectropolarimetric imaging system is proposed to acquire the spectral, polarimetric and spatial properties of the skin. After acquiring the spectropolarimetric imagery, an empirical line correction method is used to analyse the polarimetric spectra differences between normal skin and skin with a chicken pox scar. To enhance the visual difference between normal skin and skin with a chicken pox scar, and the polarimetric and spectral characteristics of the skin in different states, a false colour mapping based spectropolarimetric image fusion method is proposed that combines the intensity image, the degree of linearity of the polarization image, and the phase of the polarization image, with spectral imagery. We demonstrate experimentally that this imaging system can be used to discriminate between the different skin pathological conditions efficiently.

Image-based information is helpful for image-guided surgery therapy using medical imaging devices. In this paper, we present an image-based augmented reality model for potential medical application. Firstly, texture image is generated from two orthogonal images with multi-resolution technique. The surface of a 3D head based on MRI images is flattened onto 2D plane with cylindrical projection method. Then line-pair 2D warping method is used to determine the feature-based positional relationship between the texture image and the flattened image. The information of anatomical structure from medical images can be preserved for the future medical application. Experimental results show that the method can photo-realistically render 3D face with texture mapping. Finally, simple patient-to-model registration is used to obtain interactive augmented reality display of a surgical simulation for a simulated cyst in corpus callosum.