Efficient Prostate Tissue Classification with MobileNetV2: A Comparative Study

This paper represents a comparative analysis of the Convolutional Neural Networks (CNN), which is applied to the prostate tissue classification, pinpointing an overall accuracy of 71.79%. The study rigorously identifies multiple CNN architectures, with the focus on the MobileNetV2, to assess their efficacy in differentiating among the four categories: Benign, [Malignant] Pre-B, [Malignant] Pro-B, and [Malignant] early Pre-B. The results represent that while MobileNetV2 attains a balance between computational productivity and performance, it exhibits strengths in benign and early malignant tissue classification. However, the model undergoes challenges in accurately distinguishing between certain malignant stages, such as [Malignant] Pre-B and [Malignant] Pro-B. These results underscore the potential of CNNs in the classification of medical images but also point to the prerequisite for the further optimization and the enhancement. Future research should emphasize on finetuning architecture, incorporating the advanced data augmentation techniques, and analysing techniques to improve the classification accuracy and broadening in prostate tissue diagnosis.