Lightweight Truncated Fused-MirrorNet for Classification and Analysis of Histopathology Images.

Renal cell carcinoma is the primary cause of cancer-related mortality, highlighting the importance of early detection and accurate diagnosis. Manual histopathology image classification methods include limitations such as labor effort, time consumption, and interpathologist variations, which can lead to misdiagnosis, especially in the early stages. An autonomous solution based on deep learning is essential to overcome these constraints. However, vision-based models require significant processing resources and data sets, which pose difficulties for low-end infrastructures. In this study, we have described an approach for analyzing histopathology images using a lightweight truncated Fused-MirrorNet model. With its mirrored architecture, we use partial layer freezing and feature fusion approaches to improve performance. In kidney histopathology image analysis, our suggested strategy outperforms existing CNN and vision transformer models in histopathology image classification. We significantly reduced training time while preserving classification accuracy. The proposed model is deployable, scalable, and reproducible, allowing it to be used on low-end devices. Our strategy also makes it easier to create vision-based deep learning models by removing the requirement for sophisticated computational methodologies and procedures. The proposed model and the comparison models were trained and evaluated using histopathology images from two datasets. The experimental results reveal that the proposed model (Fused-MirrorNet) surpassed the performance of state-of-the-art models used for the classification of histopathology images. The proposed model achieves an accuracy of 92.60% and 90.00% in the TCGA kidney and BreakHis datasets, respectively. These findings indicate that the research conducted to develop the suggested model produced favorable outcomes.