Boston CCTD Task: Cognitive Impairment Classification

This model classifies patient transcripts based on the Boston Cookie Theft Picture Description Task (CCTD). In this assessment, subjects are asked to describe the events occurring in the standardized image below. The model analyzes the linguistic patterns, vocabulary, and syntax within their spoken transcripts to differentiate between Cognitively Normal and Impaired individuals.

Project Details & Methodology

• Model Architecture: Utilized DistilBertTokenizerFast and DistilBertForSequenceClassification.
• Training Source: Training code was adopted from PROCESS-2/codes/PROCESS2_BASELINE_LLM.py.
• Dataset: Powered by the CognoSpeak/PROCESS-2 dataset.
• Task Specifics: Implementation is limited strictly to the Cookie Theft Task (CTD).
• Classification: Performs 2-class classification, merging MCI (Mild Cognitive Impairment) and Dementia categories into a single 'Cognitively Impaired' group.
• Input Data: Utilizes manual transcripts for classification accuracy.