Attention-driven triple path residual 3D CNN for efficient brain tumour detection

Le résumé fourni par la source

Brain tumour diagnosis and treatment is a very challenging task as gliomas are most aggressive ones and sometimes lead to fatality. The main focus of this study is to classify the LLG (Low Grade Glioma) and HGG (High Grade Glioma) using MRI (Magnetic Resonance Image). To enhance the feature extraction and preprocessing we introduced a novelty i.e. MHA (Multi Head Attention) which is integrated with a three-path CNN (Convolution Neural Network) design. In each path a distinct multi-scale feature is extracted which are then refined through MHA and embedding concatenation which ensures the precise attention to critical tumour region. This proposed model is the demonstration of how it has superior performance in enhanced distinction of gliomas and overcomes the challenges like diverse image characteristics and subtle tumour boundaries. The Evaluation done on the BraTS dataset shows the improvement and achieved an impressive accuracy of 97.98% and generalizability as compared to the state-of-art models.

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