AI Stereotyped Behavior Detection for Autism Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current ASD diagnostic systems face challenges in early detection due to reliance on surveys and clinical observations, difficulty in observing stereotyped behaviors in unfamiliar environments, and limitations in assessing behavioral patterns without video data, leading to delayed diagnoses and inefficient detection of repetitive and periodic behaviors.
Innovation Solution
An AI-based integrated method for stereotyped behavior detection using a pipeline that includes object detection, periodicity detection through dense sampling and self-similarity matrices, and classification of behaviors as body or hand stereotypies, with a Vision-Language Model for object usage analysis, enabling accurate detection of stereotyped behaviors in long video sequences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If conventional ASD diagnostic systems rely on surveys and clinical observations by specialists, then the diagnosis can be conducted with existing resources, but the diagnostic time period becomes long and early diagnosis is delayed
Solution Approach 1:
The patent replaces the manual mechanical process of clinical observation and survey-taking with an automated AI-based video analysis system. The system automatically detects stereotyped behaviors through computer vision algorithms, eliminating the need for specialists to manually observe and record behaviors during limited diagnostic time, thereby significantly reducing diagnostic time while increasing automation.
Solution Approach 2:
The system enables self-service diagnosis by automatically analyzing video data captured in natural environments without requiring specialized medical professionals to be present during the observation period. The automated system processes video data, detects stereotyped behaviors, and provides diagnostic support independently, allowing early detection without waiting for specialist appointments.
2Reliability
If stereotyped behaviors are observed during limited diagnostic examination time in a hospital, then the examination process remains efficient, but the behaviors are difficult to observe due to unfamiliar examination environment and random occurrence
Solution Approach 1:
The system performs preliminary action by capturing video data in advance during natural environments before the formal diagnostic examination. Video data is collected in homes or childcare settings where children naturally exhibit stereotyped behaviors, and this pre-captured data is then analyzed during the diagnostic process, ensuring reliable detection without requiring complex real-time observation systems during limited hospital visits.
Solution Approach 2:
The patent introduces an intermediary system that bridges the gap between natural environment observation and formal diagnostic examination. The AI-based video analysis system acts as an intermediary that processes video data captured in natural settings and translates it into diagnostic information, enabling reliable detection of stereotyped behaviors without requiring direct observation during the limited diagnostic time window.
3Measurement precision
If parent interviews are conducted to assess behavioral patterns, then the diagnostic process remains simple, but objective assessment of degree and patterns is limited without video data
Solution Approach 1:
The system replaces the subjective mechanical process of parent interviews with an automated video-based objective measurement system. The AI-based analysis objectively quantifies behavioral patterns, frequency, and duration directly from video data, eliminating the limitations of subjective parent reporting while providing precise, measurable diagnostic information without requiring extensive interview time.
4Reliability
If conventional behavior detection technologies use predefined behavior classes and training data, then the system can be trained on specific behaviors, but detection performance drops rapidly when behaviors not present in training data occur in open-set settings
Solution Approach 1:
The patent applies parameter changes by transforming the detection approach from fixed predefined classes to a dynamic parameter-based detection system. The system detects stereotyped behaviors by analyzing temporal parameters such as periodicity, rhythm, and repetition patterns rather than relying on fixed behavioral categories, enabling stable detection performance across diverse and unpredictable behavior types without requiring extensive retraining.
Data Source
AI summary
Disclosed herein is a method for stereotyped behavior detection for supporting diagnosis of Autism Spectrum Disorder (ASD). The method includes detecting a target object to be assessed in an input video, detecting a section in which a periodic behavior occurs using an image sequence of the target object, and classifying a stereotyped behavior in the section in which the periodic behavior occurs.