Efficient recognition of Parkinson's disease mice on stepping characters with CNN.

Parkinson's disease (PD), as the second most prevalent neurodegenerative disorder worldwide, impacts the quality of life for over 12 million patients. This study aims to enhance the accuracy of early diagnosis of PD through non-invasive methods, with the goal of enabling earlier intervention in the disease process. To this end, we constructed an open-field environment using flexible sensors under dark conditions, conducting experiments on a mouse model of Parkinson's disease alongside a normal control group. By processing footprint images collected in the absence of light-employing numerical area summation for noise reduction, adaptive enhancement algorithms based on pixel values, and a high-accuracy Convolutional Neural Network algorithm. And integrating motion data analysis, we achieved effective fusion of footprint images and behavioral data. After comprehensive analysis using decision-level fusion techniques and a Naive Bayes classifier, the results showed that the average classification accuracy for PD mice reached 96.56% when employing a multimodal data fusion strategy, demonstrating a significant advantage over using a single image recognition technique. This research not only highlights the substantial potential of applying multimodal data fusion strategies in the early detection of Parkinson's disease but also proves that such an approach can significantly improve diagnostic accuracy. The findings provide new insights into the early diagnosis of Parkinson's disease and other neurodegenerative disorders.