Enhancing Brain Tumor Detection with CNN and Random Forest: A Balanced Approach to Accuracy and Interpretability

Résumé fourni par la source

Brain tumor classification is an important step in diagnosis and treatment and therefore correct and interpretable results are all important. In this research, a dual-architecture approach that integrates CNNs for feature extraction and RF for the classification process is presented in order to obtain excellent diagnostic accuracy while improving interpretation at the same time. The model was evaluated on five tumor types: The results for each tumor type were significantly higher than the chance level; glioma, meningioma, pituitary adenoma, metastatic tumor, and no tumor showed the maximum accuracy of 97.2% for pituitary adenoma and 96.9% for no tumor. High F1 Scores in all the classes with an F1 score of 0.96 for Pituitary Adenoma and 0.97 for No Tumor were established by the model, in addition to a high recall ratio of almost 97.8 percent thus effectively reducing the false negatives. The understandability of the framework was improved using the random forest algorithm together with feature importance and Grad-CAM for visualizing the important features and the regions of the images that the model pays attention to. These insights give clinicians a clear and accurate diagnosis enhancing of their work. These results showed that the proposed CNN+RF hybrid approach provided better performance than the pure CNN and RF models in their ability to learn an accurate and understandable model. This framework is a major advancement in AI-based tumour classification and opens the door for its application in clinical settings.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.