Automatic, Qualitative Scoring of the Clock Drawing Test (CDT) Based on U-Net, CNN and Mobile Sensor Data

Area of Science:

• Neurology
• Computer Science
• Medical Imaging

Background:

• The Clock Drawing Test (CDT) is a popular screening tool for cognitive functions, often used in diagnosing neurological diseases.
• Current CDT assessment relies on manual, paper-based methods, lacking standardization and efficiency.
• Advancements in mobile technology and deep learning necessitate automated, qualitative scoring systems for the CDT.

Purpose of the Study:

• To present mCDT, a mobile phone application for administering the Clock Drawing Test.
• To introduce a novel, automatic, and qualitative scoring method for the CDT using mobile sensor data and deep learning algorithms.
• To evaluate the performance of the mCDT scoring system in differentiating dementia disease subtypes.

Main Methods:

• Developed mCDT mobile application to capture CDT images and sensor data.
• Utilized deep learning models: Convolutional Neural Network (CNN) and U-Net for image segmentation (contour, hands, digits).
• Trained models on 159 CDT images and the MNIST database; scored parameters including contour, numbers, hands, and center.

Main Results:

• The mCDT system achieved high performance metrics across parameters: sensitivity (80.21-98.42%), specificity (86.21-95.93%), accuracy (87.21-96.80%), and precision (93.90-98.15%).
• Performance testing on 219 subjects demonstrated the system's reliability, validated by clinical experts.
• The automatic scoring system effectively segmented clock contours, hands, and digits.

Conclusions:

• The mCDT application and its deep learning-based scoring system offer a standardized, automated, and qualitative method for CDT assessment.
• The system shows significant utility in clinical practice for differentiating dementia disease subtypes.
• mCDT provides a valuable tool for cognitive function research and clinical evaluation.