Machine learning model for predicting the conversion to dementia using the Cube Copying Test

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Early dementia detection is crucial for effective management.
• Current screening methods can be burdensome for patients.
• Identifying preclinical or mild cognitive impairment (MCI) markers is essential.

Purpose of the Study:

• To develop a machine learning model for predicting dementia conversion within 3-5 years.
• To utilize Cube Copying Test (CCT) drawings for early dementia detection.
• To identify subtle drawing pattern changes indicative of future dementia.

Main Methods:

• Retrospective analysis of CCT drawings from 767 patients.
• Development of a deep learning-based anomaly detection model.
• Utilized Shapley Additive exPlanations (SHAP) for feature importance analysis.

Main Results:

• The model achieved an area under the curve (AUC) of 0.85 for dementia conversion prediction.
• PatchCore-derived features were identified as strong predictors.
• Early constructional apraxia-like symptoms were detected in drawings of converters.

Conclusions:

• Deep learning models can detect early drawing distortions in preclinical/MCI stages.
• These distortions differ from normal aging patterns.
• This approach can significantly improve dementia conversion prediction accuracy.