Automated grading of renal cell carcinoma using an efficient deep learning approach on histopathology images

In the healthcare industry, Renal Cell carcinoma (RCC) is the most common form of kidney cancer, and it is heterogeneous with intrinsic histopathological features. Moreover, grading RCC from the histopathology slides is essential for prognosis and determination of treatment strategy, but manual grading is labor intensive, time consuming and requires addressing inherent observer variability. Recently, Deep Learning (DL) approaches have shown significant advantages in medical image analysis, However, applying DL approaches to RCC grading still faces several challenge, which include small, annotated datasets and algorithms that can adequately represent intricate histopathological features. Besides, the proposed study provides datasets of digitalized histopathology scans of RCC slides, which are annotated by expert pathologists that are sufficient for training and testing models. Therefore, the proposed Dual Attention-Gated Dilated MobileNetV2 (DAG-DMobileNetV2) is employed for automated RCC grading. The model addresses these challenges by using dilated convolutions to increase the receptive field and generate a larger context without a reduction in resolution, hence allowing the model to extract both fine-grained and global features effectively. In addition, the dual attention mechanism, composed of channel and spatial attention mechanisms that enables the model to dynamically emphasize important areas and feature channels while suppressing irrelevant noise and concentrating on diagnostically important regions. Thus, the MobileNet backbone enables the model to operate efficiently, allowing for its suitability for deployment in real-time clinical settings. Hence, the contributions of the proposed model are the distinct attention modules coupled with dilated convolutions, allowing for accurate discrimination of the 4 different grades of RCC. Thus, the proposed DAG-DMobileNetV2 achieves improved classification performance, improved feature representation, and decreased computational complexity resulting in faster inference. Moreover, the attention mechanisms for DAG-DMobileNetV2 should automatically accommodate and adapt to variability in histopathological image quality.