Research on multi-label recognition of tongue features in stroke patients based on deep learning.

Stroke has become the leading cause of disability in adults worldwide. Early precise rehabilitation intervention is crucial for the recovery of stroke patients, with the key lying in accurately identifying patients' physical characteristics during the rehabilitation phase. Compared to diagnostic techniques such as medical neuroimaging, traditional Chinese medicine(TCM) tongue diagnosis offers good accessibility and ease of application. However, conventional TCM tongue diagnosis relies on the experience of doctors, which introduces a degree of subjectivity, especially since stroke patients exhibit unique characteristics in tongue texture, shape, and coating, making accurate diagnosis more challenging. To address this issue, this paper proposes a deep learning-based automatic recognition approach for the tongue images of stroke patients, aiming to improve the accuracy of automatic extraction and recognition of stroke-related tongue features through image processing and machine learning techniques. First, this study performs image cropping and data augmentation on tongue images. Then, considering that tongue color, coating color, and coating texture are interrelated in TCM theory and jointly reflect the body's physiological and pathological state, a label-guided multi-label recognition model for tongue images is designed. This model extracts features from the tongue images of stroke patients, learns the correlations among the features, and performs classification to automatically identify key characteristics such as tongue shape, color, and coating. Finally, the model's performance is quantitatively evaluated. Experimental results show that the proposed deep learning model outperforms several advanced deep learning models, such as resnet and densenet, and existing single-task tongue classification models in automatically recognizing stroke patients' tongue images. This research improves the accuracy of feature extraction and recognition of tongue characteristics in stroke patients during rehabilitation, providing a convenient and feasible technical approach for real-time evaluation and diagnosis in the stroke rehabilitation process. It has significant clinical application value and research significance.