Abstract:Importance: Dystonia in cerebral palsy (CP) is common and debilitating but under-diagnosed, particularly in the legs. Only 13% of children with CP and leg dystonia have that dystonia diagnosed in clinic. Yet, early and accurate dystonia diagnosis is critical for appropriate treatment selection. Objective: To facilitate automated leg dystonia diagnosis in CP, our objective was to develop open-source software that used machine learning (ML) models and clinically available video data. Design, Setting, and Participants: In this cross-sectional study, videos of children with CP (age 5 and older) doing a seated alternating hand open-close task were acquired between 1/1/2020 and 12/14/23 from two tertiary care CP centers (193 videos from Center 1, 30 videos from 2). These videos were assessed for leg dystonia by 8 pediatric movement disorders physicians (gold-standard diagnosis by expert consensus). Movement features they cited when identifying dystonia were extracted from their consensus-building discussions using conventional content analysis and were then quantified from the videos leveraging our open-source software. Quantified features from 163 videos from Center 1 were used to train ML models (4664 models based on 11 ML algorithms) which were then tested on 30 videos from Center 1 and 30 videos from Center 2. Main Outcome(s) and Measure(s): Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and accuracy of ML models were computed as compared to gold-standard leg dystonia diagnosis by expert consensus. Results: Expert-cited quantifiable movements associated with dystonia diagnosis were used to identify 63 kinematics features to extract from videos. Feature importance analysis showed that variance features had higher importance than amplitude features. ML models achieved 84% sensitivity, 84% specificity, 84% PPV, 85% NPV, and 84% accuracy for identifying leg dystonia in testing videos from Center 1 and 100% sensitivity, 67% specificity, 75% PPV, 100% NPV, and 83% accuracy for identifying leg dystonia in videos from Center 2. However, for videos where at least 60% of experts agreed regarding the presence or absence of dystonia (n=23/30 from Center 2), ML models were 100% accurate for identifying leg dystonia. Conclusions/Significance: We have developed open-source software that can use clinically acquired videos to identify leg dystonia in children with CP. These methods may be useful for facilitating clinical dystonia diagnosis in children with CP.

Deep Learning-Based Classification of Hyperkinetic Movement Disorders in Children

A Novel Video-Based Methodology for Automated Classification of Dystonia and Choreoathetosis in Dyskinetic Cerebral Palsy During a Lower Extremity Task

Automated Leg Dystonia Diagnosis in Children with Cerebral Palsy Using Machine Learning

Deep Learning for the Assisted Diagnosis of Movement Disorders, Including Isolated Dystonia

Head movement dynamics in dystonia: a multi-centre retrospective study using visual perceptive deep learning

Home-Based Measurements of Dystonia in Cerebral Palsy Using Smartphone-Coupled Inertial Sensor Technology and Machine Learning: A Proof-of-Concept Study

Detecting Daily Living Gait Amid Huntington's Disease Chorea using a Foundation Deep Learning Model

Bradykinesia Recognition in Parkinson’s Disease Via Single RGB Video

Classification of Abnormal Hand Movement for Aiding in Autism Detection: Machine Learning Study

Motor assessment of X-linked dystonia parkinsonism via machine-learning-based analysis of wearable sensor data

Automatic two-dimensional & three-dimensional video analysis with deep learning for movement disorders: A systematic review

Action-Based ADHD Diagnosis in Video

Interpretable video-based tracking and quantification of parkinsonism clinical motor states

Video-based diagnosis support system for pianists with Musician's dystonia

Dynamical system based compact deep hybrid network for classification of Parkinson disease related EEG signals

A Clinically Guided Graph Convolutional Network for Assessment of Parkinsonian Pronation-Supination Movements of Hands

A Spatio-temporal Attention-based Model for Infant Movement Assessment from Videos

Deep neural network technique for automated detection of ADHD and CD using ECG signal

Evaluation of Three Machine Learning Algorithms for the Automatic Classification of EMG Patterns in Gait Disorders

Automated Movement Detection with Dirichlet Process Mixture Models and Electromyography

Applying Machine Learning to Kinematic and Eye Movement Features of a Movement Imitation Task to Predict Autism Diagnosis