Histopathological image classification using dilated residual grooming kernel model

Breast cancer is being diagnosed earlier and more accurately through deep learning and machine learning models. This study contributes to medical science and technology using a novel deep learning-based model for detecting tiny cancer cells and accurately diagnosing cancer cells. The proposed approach uses strain normalisation to resolve colour divergence concerns in the breast cancer histopathological image classification (BreakHis) and breast cancer histopathological annotation and diagnosis (BreCaHAD) image datasets. Followed by 19 various parameters such as scaling, rotation, flip, resize, and gamma value are used to tackle the overfitting problem. The proposed dilated residual grooming kernel (DRGK) model is a 19-layer model and contains a proposed multi-scale dilated convolution (MSDC) unit to extract the features efficiently by detection of tiny objects and thin boundaries without adding complexity. Along with MSDC unit and convolution, pooling, down sampling, the proposed DRGK model accurately detects the cancer cells with accuracy of 98.5%.