Classification of Acute Radiation Dermatitis by Active-Guided-Based T-S Fuzzy Classifier

Acute radiation dermatitis (AD) is a prevalent side effect during radiotherapy that, if not intervened in a timely manner, can reduce patients' quality of life, impact their health, and even result in mortality. Therefore, a highly precise hierarchical TSK fuzzy classifier is devised to cater to the necessity for early intervention in AD. However, the existing hierarchical TSK fuzzy classifiers grapple with the issue of arbitrary input-output projection between adjacent layers and lacks actively guided learning methods. This discrepancy results in relatively substantial errors in the hierarchical learning process, hindering the implementation of targeted active learning. Consequently, this study introduces an innovative labeled actively guided hierarchical Takagi-Sugeno-Kang (TSK) fuzzy classifier (LAG-TSK-FC). The classifier leverages optimized zero-order TSK fuzzy classifiers and is constructed upon an enhanced cascaded structure. Additionally, this study proposes a technique involving label intervention for error amplification to enhance the accuracy of each layer, thereby ensuring classification accuracy and generalization performance. Furthermore, an optimized approach for minimizing training error to compute the consequent parameters is presented. Through label guidance, intentional consistency is achieved between the preceding and succeeding layers, ensuring the transmission of the same physical interpretation to maintain high accuracy and interpretability of the classifier. Finally, experimental validation on four AD datasets demonstrates that LAG-TSK-FC exhibits commendable classification and generalization performance, maintaining high accuracy.