Using partial decision trees to predict Parkinson’s symptoms: A new approach for diagnosis and therapy in patients suffering from Parkinson’s disease

doi:10.1016/j.compbiomed.2011.11.008

Computers in Biology and Medicine

Volume 42, Issue 2, February 2012, Pages 195-204

https://doi.org/10.1016/j.compbiomed.2011.11.008 Get rights and content

Abstract

In this work we present a method based on partial decision trees and association rules for the prediction of Parkinson's disease (PD) symptoms. The proposed method is part of the PERFORM system. PERFORM is used for the treatment of PD patients and even advocate specific combinations of medications. The approach presented in this paper is included in the data miner module of PERFORM. A patient performs some initial examinations and the module predicts the future occurrence of the symptoms based on the initial examinations and medications taken. Using the method, the expert can prescribe specific medications that will not cause, or postpone the appearance of specific symptoms to the patient. The approach employed is able to provide interpretation for the predictions made, by providing rules. The models have been developed and evaluated using real patient's data and the respective results are reported. Another functionality of the data miner module is the extraction of rules through a user friendly interface using association rule mining algorithms. These rules can be used for the prediction analysis of patient's reaction to certain treatment plans. The accuracy of the symptoms' prediction ranges from 57.1 to 77.4%, depending on the symptom.

Introduction

Parkinson's disease is a complex disease which affects the patient's movement. Initial symptoms include tremor, muscle rigidity and slowness in movements. Symptoms are controlled with suitable medication. The aim of the therapy is to alleviate disease's symptoms while alleviating also movement disorders caused by the therapy itself.

Over the past decades various methodologies and systems have been proposed for the monitoring and assessment of Parkinson's disease (PD) patients. However, only one proposes treatment plans [1]. Two categories of methodologies can be distinguished. The first category focuses on analyzing specific motion tests and assessing the patient motion status periodically ([1], [2], [3], [4], [5], [6], [7], [8], [9]). The second category focuses on the study of specific symptoms during daily patient activities, through the use of different types of wearable movement sensors ([10], [11], [12], [13], [14], [15]). However, only a few of them can potentially support treatment plan suggestions ([12], [13], [15]).

Decision support systems based on data mining techniques have been developed in a wide variety of application domain, both in medicine and biology. The usage of data mining technologies allows the development of decision support systems in domain where no prior knowledge exists, but there are data available [16], [17], [18], [19], [20]. The general concept of these studies is the extraction of knowledge and rules from a set of data and the development of decision support systems for diagnosis, prediction and classification. For example, in [16], [17], they have developed a decision support system for blood donation and transfusion service from patient blood samples. In [18], decision trees were employed for prostate cancer diagnosis. In [19], different decision tree induction methodologies were used for determining recurrence-free survival of breast cancer patients. In [20], [21] association rules were employed for hybrid medical image classification and diagnostic analysis of patients with hypertension.

In this work we present the data miner module of the PERFORM system. PERFORM studies specific symptoms during daily PD patient activities, using a set of wearable sensors, but additionally incorporates the patient's disease profile and previous measurements in order to suggest appropriate treatment plans. PERFORM is designed to support both patients and clinicians in the daily management of chronic neurodegerative diseases [22]. The PERFORM system provides an integrated platform for the monitoring, modeling and management of the PD patient health status. More specifically, PERFORM functionalities include:

•
All day patient monitoring inside and outside the home,
•
Detection and quantification of patient symptoms,
•
Patient status assessment,
•
Disease evolution assessment.

PERFORM aspires to change the way by which health services are delivered today and support the growing need of personalized treatment of PD patients. As the European population becomes older, age related diseases will demand more healthcare resources. PERFORM can support clinicians and hospitals in treating the increasing number of neurodegenerative patients and to improve the quality of the provided health services. In fact, its usage can improve patient status assessment, as it proposes an objective method in contrary to the current subjective clinical practice, which is based on the patient-clinician communication over periodic patient visits in the hospital [23].

The data miner module of PERFORM, predicts the occurrence of symptoms in patients, based on initial examinations and prescribed drugs. Using the proposed method, the expert can prescribe specific medications to PD patients, tailored to the specific patient needs. PD patients are able to receive medications that will not cause, or will cause harmelss symptoms to them. The method is able to predict the future occurrence of 15 different PD symptoms. PD patients are able to know in advance symptoms that will occur to them and with the help of the medical experts define ways for diminishing or ameliorating them. The predictive models can provide interpretation for the decisions made to the expert during, since they are based on sets of rules. Another functionality is the discovery of new knowledge from a database of patients suffering from PD. This functionality can be used to study patient's reactions to treatment plans.

Section snippets

Methods

PERFORM consists of three subsystems: the wearable multi-sensor monitor unit (wearable sensors), the local base unit and the centralized hospital unit [24]. The PERFORM Wearable Device or Multi-Sensor Monitor Unit is physically located in the patient's setting along with the local base unit (LBU). It is mainly used to monitor patient's daily motor activity and status. The signals recorded through the various sensors are later transferred to the LBU where they are processed. The Wearable Device

Dataset

For the development and evaluation of the data miner predictive models, 230 patients with confirmed PD from the neurological clinic of the University Hospital of Ioannina have been enrolled. From those patients, the physicians kept records for the time they have been diagnosed with PD, along with the respective symptoms, and records of later visits, along with the symptoms that the patients presented then. Moreover, medications prescribed at any stage of therapy have been taken into

Results

Below we present the symptom prediction models generated for every symptom, which is represented as different classification task. The prediction of each rule is presented after “:”. “1” denotes the occurrence of the symptom, while “0” means that the symptom will not probably occur. The rules that are generated from the predictor are applied in order to predict the future occurrence of symptoms in patients. The following rules are used serially, from left to right. If none of the rules apply,

Discussion

We have presented a method based on partial decision trees and association rules for building predictive models for PD symptoms and for discovering new knowledge for PD symptoms in the form of association rules. The approach is based on wrapper feature selection and classification rule discovery based on partial decision trees algorithms. Real data have been acquired for the development and the evaluation of the predictive models. A great advantage of the proposed work is that it deals with all

Conclusions

We have presented the data miner module of the PERFORM system for developing PD symptom prediction models and for discovering new knowledge in the form of association rules. The module reported promising results, based on an initial set of PD patients, which can be further enhanced using additional patient records to address the imbalance problem. To fully reveal the potential of such a system, it needs to be applied in real world clinical settings and in real every day clinical practice.

Conflict of interest statement

None

Acknowledgments

This work is partly funded by the European Commission (project PERFORM–A sophisticated multi-parametric system for the continuous-effective assessment and monitoring of motor status in Parkinson's disease and other neurodegenerative diseases: FP7-ICT-2007-1-215952).

References (34)

M. Caligiuri et al.
Quantitative measurement of handwriting in the assessment of drug-induced parkinsonism
Hum. Movment Sci.
(2006)
M.I. Chelaru et al.
Levodopa-induced dyskinesias detection based on the complexity of involuntary movements
J. Neurosc. Methods
(2010)
R.A. Joundi et al.
Rapid tremor frequency assessment with the iPhone accelerometer,
Parkinsonism Relat. Disord.
(2011)
B.N. Li et al.
On decision making support in blood bank information systems
Expert Syst. Appl.
(2008)
B.N. Li et al.
SIBAS: A blood bank information system and its 5-year implementation at Macau
Comput. Biol. Med.
(2007)
M. Ture et al.
Using Kaplan-Meier analysis together with decision tree methods (C&RT, CHAID, QUEST, C4.5 and ID3) in determining recurrence-free survival of breast cancer patients
Expert Syst. Appl.
(2009)
J. Westin et al.
A fuzzy rule-based decision support system for Duodopa treatment in Parkinson's disease
Eur. J. Neurol.
(2006)
D. Allen et al.
On the use of low-cost computer peripherals for the assessment of motor dysfunction in Parkinson's disease—quantification of bradykinesia using target tracking tasks
IEEE Trans. Neural Syst. Rehabil. Eng.
(2007)
B. Christe et al.
Clinical assessment of motor function: a processes oriented instrument based on a speed-accuracy trade-off paradigm
Behav. Neurol.
(2007)
J. Gourb et al.
Movement patterns of peak-dose levodopa-induced dyskinesias in patients with Parkinson's disease
Brain Res. Bull.
(2007)

J. Giuffrida et al.

ParkinSense comparison to the unified Parkinson's disease rating scale: preliminary tremor and bradykinesia results

Movement Disord

(2007)

C. De Luca, S. Roy, S. Chang, and S. Nawab, A wearable monitor for movement disorders in Parkinson's disease, Annual...

L. Fuyuan et al.

Multi-resolution entropy analysis of gait symmetry in neurological degenerative diseases and amyotrophic lateral sclerosis

Med. Eng. Phys.

(2008)

S. Patel et al.

Monitoring motor fluctuations in patients with Parkinson's disease using wearable sensors

IEEE Trans. Inf. Technol. Biomed.

(2009)

N. Keijsers et al.

Ambulatory motor assessment in Parkinson's disease,

Movment Disord.

(2006)

A. Salarian et al.

Quantification of tremor and bradykinesia in Parkinson's disease using a novel ambulatory monitoring system

IEEE Trans. Biomed. Eng.

(2007)

D. Zwartjes et al.

Development of a system for measurement and analysis of tremor usinga three-axis accelerometer,

IEEE Trans. Biomed. Eng.

(2010)

Cited by (28)

Early diagnosis of Parkinson's disease: A combined method using deep learning and neuro-fuzzy techniques
2023, Computational Biology and Chemistry
Citation Excerpt :
On the other hand, in terms of specificity, LR presented the best results. Exarchos et al. (2012) used the DT module and association rules for PD diagnosis. The designed approach was a portion of the PERFORM system, which was utilized for PD controlling and therapy.
Predicting Unified Parkinson's Disease Rating Scale (UPDRS) in Total- UPDRS and Motor-UPDRS clinical scales is an important part of controlling PD. Computational intelligence approaches have been used effectively in the early diagnosis of PD by predicting UPDRS. In this research, we target to present a combined approach for PD diagnosis using an ensemble learning approach with the ability of online learning from clinical large datasets. The method is developed using Deep Belief Network (DBN) and Neuro-Fuzzy approaches. A clustering approach, Expectation-Maximization (EM), is used to handle large datasets. The Principle Component Analysis (PCA) technique is employed for noise removal from the data. The UPDRS prediction models are constructed for PD diagnosis. To handle the missing data, K-NN is used in the proposed method. We use incremental machine learning approaches to improve the efficiency of the proposed method. We assess our approach on a real-world PD dataset and the findings are assessed compared to other PD diagnosis approaches developed by machine learning techniques. The findings revealed that the approach can improve the UPDRS prediction accuracy and the time complexity of previous methods in handling large datasets.
Remote tracking of Parkinson's Disease progression using ensembles of Deep Belief Network and Self-Organizing Map
2020, Expert Systems with Applications
Citation Excerpt :
It is caused when the level of dopamine drops by about 70%. The main symptoms of PD include tremor, muscle stiffness, bradykinesia (motor features) as well as symptoms such as constipation, cardiac arrhythmia and sleep disorders (non-motor features) (Che et al., 2017; Engelender & Isacson, 2017; Exarchos et al., 2012; Graça, e Castro, & Cevada, 2014; Jankovic, 2008). Most PD patients suffer from difficulties in walking and speaking or in performing routine tasks.
Parkinson’s Disease (PD) is one of the most prevalent neurological disorders characterized by impairment of motor function. Early diagnosis of PD is important for initial treatment. This paper presents a newly developed method for application in remote tracking of PD progression. The method is based on deep learning and clustering approaches. Specifically, we use the Deep Belief Network (DBN) and Support Vector Regression (SVR) to predict Unified Parkinson's Disease Rating Scale (UPDRS). The DBN prediction models were developed by different epoch numbers. We use a clustering approach, namely, Self-Organizing Map (SOM), to improve the accuracy and scalability of prediction. We evaluate our method on a real-world PD dataset. In all, nine clusters were detected from the data with the best SOM map quality for clustering, and for each cluster, a DBN was developed with a specific number of epochs. The results of the DBN prediction models were integrated by the SVR technique. Further, we compare our work with other supervised learning techniques, SVR and Neuro-Fuzzy techniques. The results revealed that the hybrid of clustering and DBN with the aid of SVR for an ensemble of the DBN outputs can make relatively better predictions of Total-UPDRS and Motor-UPDRS than other learning techniques.
Prediction of irinotecan toxicity in metastatic colorectal cancer patients based on machine learning models with pharmacokinetic parameters
2019, Journal of Pharmacological Sciences
Citation Excerpt :
Thus, the clinician is able to anticipate to any of the studied toxicities and make decisions accordingly; both to treat such toxicities and to adjust the dosing for subsequent cycles, in combination with additional information derived from pharmacogenetics43 and pharmacokinetics.44 This work is limited by the size of the population sample and, although the employed validation techniques have been applied in other published works with accredited reliability when the available cohort is small,21,23–25,30 this fact should be considered when applying these models in clinical practice. Hence, the adequacy of the algorithms to the target population must be checked prior to their application.
Irinotecan (CPT-11) is a drug used against a wide variety of tumors, which can cause severe toxicity, possibly leading to the delay or suspension of the cycle, with the consequent impact on the prognosis of survival. The main goal of this work is to predict the toxicities derived from CPT-11 using artificial intelligence methods.
The data for this study is conformed of 53 cycles of FOLFIRINOX, corresponding to patients with metastatic colorectal cancer. Supported by several demographic data, blood markers and pharmacokinetic parameters resulting from a non-compartmental pharmacokinetic study of CPT-11 and its metabolites (SN-38 and SN-38-G), we use machine learning techniques to predict high degrees of different toxicities (leukopenia, neutropenia and diarrhea) in new patients.
We predict high degree of leukopenia with an accuracy of 76%, neutropenia with 75% and diarrhea with 91%. Among other variables, this study shows that the areas under the curve of CPT-11, SN-38 and SN-38-G play a relevant role in the prediction of the studied toxicities.
The presented models allow to predict the degree of toxicity for each cycle of treatment according to the particularities of each patient.
Key Hospital Anxiety and Depression Scale (HADS) items associated with DSM-IV depressive and anxiety disorder 12-months post traumatic brain injury
2018, Journal of Affective Disorders
Anxiety and depression are common problems following traumatic brain injury (TBI), warranting routine screening. Self-report rating scales including the Hospital Anxiety and Depression Scale (HADS) are associated with depression and anxiety diagnoses in individuals with TBI. The relationship between individual HADS symptoms and structured clinical interview methods (SCID) requires further investigation, particularly in regard to identifying a small number of key items that can potentially be recognised by clinicians and carers of individuals with TBI.
138 individuals sustaining a complicated-mild to severe TBI completed the HADS, and the Structured Clinical Interview for DSM-IV, Research Version (SCID) at 12-months post-injury. The associations between individual HADS items, separately and in combination, as well as overall depression and anxiety subscale scores, and SCID-diagnosed depressive and anxiety disorders were analysed.
CART (Classification and Regression Tree) analysis found HADS depression item 2 “I still enjoy the things I used to enjoy” and a combination of two anxiety items, 3 “I get a sort of frightened feeling as if something awful is about to happen” and 5 “worrying thoughts go through my mind”, performed similarly to total depression and anxiety subscales in terms of their association with depressive and anxiety disorders respectively, at 12-months post-injury.
Patients were predominantly injured in motor vehicle accidents and received comprehensive care within a no-fault accident compensation system and so may not be representative of the wider TBI population.
Although validation is required, a small number of self-report items are highly associated with 12-month post-injury diagnoses.
Gait and tremor investigation using machine learning techniques for the diagnosis of Parkinson disease
2018, Future Generation Computer Systems
Citation Excerpt :
Tucker et al. [12] used remote data mining based methodology to differentiate between “on” and “off” medication states among PD patients using low-cost, non-wearable sensor hardware. Data miner module is prepared using association rule mining algorithm and classification is done by using partial decision tree by Exarchos T.P. et al. [13]. N. Tahir et al. [14] used gait parameters features for classifying gait input using support vector machine and artificial neural network.
Parkinson’s disease (PD) is a chronic and progressive movement disorder affecting patients in large numbers throughout the world. As PD progresses, the affected person is unable to control movement normally. Individuals affected by Parkinson’s disease exhibit notable symptoms like gait impairments and tremor occurrences during different stages of the disease. In this paper a novel approach has been proposed to diagnose PD using the gait analysis, that consists of the gait cycle, which can be broken down into various phases and periods to determine normative and abnormal gait. Initially, the raw force data obtained from physionet database was filtered using a Chebyshev type II high pass filter with a cut-off frequency 0.8 Hz to remove noises arising from the changes in orientation of the subject’s body and other factors during measurement. The filtered data was used for extracting various gait features using the peak detection and pulse duration measuring techniques. The threshold values of the gait detection algorithm were tuned to individual subjects. From the peak detection algorithm, various kinetic features including the heel and toe forces, and their normalized values were obtained. The pulse duration algorithm was developed to extract different temporal features including the stance and swing phases, and stride time. Tremor is a common symptom in PD. Tremor is an involuntary movement of body parts. At first the tremor may appear in a specific body part like an arm, leg or one side of the body and later it may spread to both sides . This rest tremor is a cardinal sign of PD. An average accuracy of 92.7% is achieved for the diagnosis of PD from gait analysis and tremor analysis is used for knowing the severity of PD.
A review of fuzzy cognitive maps in medicine: Taxonomy, methods, and applications
2017, Computer Methods and Programs in Biomedicine
A high percentage of medical errors, committed because of physician's lack of experience, huge volume of data to be analyzed, and inaccessibility to medical records of previous patients, can be reduced using computer-aided techniques. Therefore, designing more efficient medical decision-support systems (MDSSs) to assist physicians in decision-making is crucially important. Through combining the properties of fuzzy logic and neural networks, fuzzy cognitive maps (FCMs) are among the latest, most efficient, and strongest artificial intelligence techniques for modeling complex systems. This review study is conducted to identify different FCM structures used in MDSS designs. The best structure for each medical application can be introduced by studying the properties of FCM structures.
This paper surveys the most important decision- making methods and applications of FCMs in the medical field in recent years. To investigate the efficiency and capability of different FCM models in designing MDSSs, medical applications are categorized into four key areas: decision-making, diagnosis, prediction, and classification. Also, various diagnosis and decision support problems addressed by FCMs in recent years are reviewed with the goal of introducing different types of FCMs and determining their contribution to the improvements made in the fields of medical diagnosis and treatment.
In this survey, a general trend for future studies in this field is provided by analyzing various FCM structures used for medical purposes, and the results from each category.
Due to the unique specifications of FCMs in integrating human knowledge and experience with computer-aided techniques, they are among practical instruments for MDSS design. In the not too distant future, they will have a significant role in medical sciences.

View all citing articles on Scopus

View full text

Using partial decision trees to predict Parkinson’s symptoms: A new approach for diagnosis and therapy in patients suffering from Parkinson’s disease

Abstract

Introduction

Section snippets

Methods

Dataset

Results

Discussion

Conclusions

Conflict of interest statement

Acknowledgments

Hum. Movment Sci.

J. Neurosc. Methods

Parkinsonism Relat. Disord.

Expert Syst. Appl.

Comput. Biol. Med.

Expert Syst. Appl.

A fuzzy rule-based decision support system for Duodopa treatment in Parkinson's disease

Eur. J. Neurol.

On the use of low-cost computer peripherals for the assessment of motor dysfunction in Parkinson's disease—quantification of bradykinesia using target tracking tasks

IEEE Trans. Neural Syst. Rehabil. Eng.

Clinical assessment of motor function: a processes oriented instrument based on a speed-accuracy trade-off paradigm

Behav. Neurol.

Movement patterns of peak-dose levodopa-induced dyskinesias in patients with Parkinson's disease

Brain Res. Bull.

ParkinSense comparison to the unified Parkinson's disease rating scale: preliminary tremor and bradykinesia results

Movement Disord

Multi-resolution entropy analysis of gait symmetry in neurological degenerative diseases and amyotrophic lateral sclerosis

Med. Eng. Phys.

Monitoring motor fluctuations in patients with Parkinson's disease using wearable sensors

IEEE Trans. Inf. Technol. Biomed.

Ambulatory motor assessment in Parkinson's disease,

Movment Disord.

Quantification of tremor and bradykinesia in Parkinson's disease using a novel ambulatory monitoring system

IEEE Trans. Biomed. Eng.

Development of a system for measurement and analysis of tremor usinga three-axis accelerometer,

IEEE Trans. Biomed. Eng.