Pictorial Language Variations Analysis On Spiral Drawings For Parkinson’s Disease With Xai

Parkinson's disease (PD) is one of the most common neurodegenerative disorders. While there is no cure for this illness, early detection is crucial for slowing its progression. Digital spiral drawings tests have emerged as an effective method to evaluate this illness as they capture multiple characteristics present in pictorial languages that are useful for PD detection. However, cultural variations in writing habits could influence these features, limiting the generalizability of AI models. In this paper, three datasets of spiral drawings from different cultural backgrounds were analyzed and used to train multiple machine learning models. Eight ML models were evaluated with LightGBM achieving the best performance with a precision and recall of 93% on the three databases combined, and over 90% on precision and recall in each dataset separately. The model adapted to each dataset, with SHAP revealing varying feature importance depending on the population, showing the importance of using multi-cultural datasets for building generalized, high performing models.