Leveraging EfficientNetB5 for Accurate Classification of Diverse Human Cancer Tissues

The paper investigates the use of a fine-tuned EfficientNetB5 model for human cancer tissue sample categorizing. From Kaggle ten thousand high-resolution images across nine categories: ADI (Adipose), Back (Background), Deb (Debris), LYM (Lymphocytes), MUC (Mucus), MUS (Muscle), NORM (Normal), STR (Stroma), and TUM (Tumour). The model was improved via transfer learning using pre-trained weights and customized for the specific objective of cancer tissue classification. Among the techniques of data augmentation applied to increase generalizing power of the model and reduce overfitting were rotation, flips, and colour changes. The fine-tuned EfficientNetB5 model showed useful general accuracy of 90% in differentiating several tissue types. Especially in differentiating malignant (TUM) and non-cancerous (NORM) tissues, performance measures including precision (0.91), recall (0.89), and F1-score (0.90) verified the model's remarkable performance. The confusion matrix analysis showed that the model could correctly categorize tumour and normal tissues, but the proof of good discrimination between multiple tissue types exposed This work highlights the prospects of using deep learning, especially EfficientNetB5, in automating cancer tissue classification, so providing a realistic approach to raise diagnosis accuracy and efficiency in cancer pathology. Future research will concentrate on improving the model and customizing it to certain cancer forms thereby increasing its relevance in therapeutic environments.