Video-based evaluation system for tic action in Tourette syndrome: modeling, detection, and evaluation.

Behavioral ratings based on clinical observations are still the gold standard for screening, diagnosing, and assessing outcomes in Tourette syndrome. Detecting tic symptoms plays an important role in patient treatment and evaluation; accurate tic identification is the key to clinical diagnosis and evaluation. In this study, we proposed a tic action detection method using face video feature recognition for tic and control groups. Through facial ROI extraction, a 3D convolutional neural network was used to learn video feature representations, and multi-instance learning anomaly detection strategy was integrated to construct the tic action analysis and discrimination framework. We applied this tic recognition framework in our video dataset. The model evaluation results achieved average tic detection accuracy of 91.02%, precision of 77.07% and recall of 78.78%. And the tic score curve with postprocessing provided information of how the patient's twitches change over time. The detection results at the individual level indicated that our method can effectively detect tic actions in videos of Tourette patients without the need for fine labeling, which is significant for the long-term evaluation of patients with Tourette syndrome.

A method for the early prediction of chronic diseases based on short sequential medical data.

Noncommunicable diseases (NCDs) have become the leading cause of death worldwide. NCDs' chronicity, hiddenness, and irreversibility make patients' disease self-awareness extremely important in disease control but hard to achieve. With an accumulation of electronic health record (EHR) data, it has become possible to predict NCDs early through machine learning approaches. However, EHR data from latent NCD patients are often irregularly sampled temporally, and the data sequences are short and imbalanced, which prevents researchers from fully and effectively using such data. Here, we outline the characteristics of typical short sequential data for NCD early prediction and emphasize the importance of using such data in machine learning schemes. We then propose a novel NCD early prediction method: the short sequential medical data-based early prediction method (SSEPM). The SSEPM network contains two stacked subnetworks for multilabel enhancement. In each subnetwork, long short-term memory (LSTM) and attention layers are implemented to extract both temporal and nontemporal embedded features. During training, with prior clinical knowledge of the NCD characteristics, a random connection (RC) process is proposed for data augmentation. Comparative experiments involving ten-fold cross-validation are performed with real-world medical data to predict 5 NCDs. The result shows that the SSEPM outperforms the state-of-the-art NCD early prediction algorithms and works well in dealing with short sequential data. The results also suggest that the direct use of short sequential data could be more effective than formatting datasets with temporal exclusion limitations.

Tic Detection in Tourette Syndrome Patients Based on Unsupervised Visual Feature Learning.

A clinical diagnosis of tic disorder involves several complex processes, among which observation and evaluation of patient behavior usually require considerable time and effective cooperation between the doctor and the patient. The existing assessment scale has been simplified into qualitative and quantitative assessments of movements and sound twitches over a certain period, but it must still be completed manually. Therefore, we attempt to find an automatic method for detecting tic movement to assist in diagnosis and evaluation. Based on real clinical data, we propose a deep learning architecture that combines both unsupervised and supervised learning methods and learns features from videos for tic motion detection. The model is trained using leave-one-subject-out cross-validation for both binary and multiclass classification tasks. For these tasks, the model reaches average recognition precisions of 86.33% and 86.26% and recalls of 77.07% and 78.78%, respectively. The visualization of features learned from the unsupervised stage indicates the distinguishability of the two types of tics and the nontic. Further evaluation results suggest its potential clinical application for auxiliary diagnoses and evaluations of treatment effects.