An ensemble of deep CNNs for automatic grading of breast cancer in digital pathology images

Histopathological diagnosis is the mainstay of present-day preventive medical care service to guide the therapy and treatment of breast cancer at an early stage. Manual examination of histologic data based on clinicians’ subjective knowledge is a time-consuming, labour-intensive, and costly method that necessitates clinical intervention and competence for a fair decision. In the recent work, we have developed an ensemble of five deep CNNs to classify three grades of breast cancer using quantitative image-based assessment of digital pathology slides without any manual intervention. To produce final predictions on the dataset, a fuzzy ranking algorithm is used. On the Databiox dataset, the suggested model attained an accuracy of 79%, 75%, 89%, and 82% at 4×, 10×, 20×, and 40× magnification, respectively. Furthermore, it has been observed that the stain-normalization strategy improves the model’s classification performance on the histopathological images. In this case, the Macenko stain-normalization technique is employed which further enhances the performance of the proposed ensemble model up to 80%, 100%, 100%, and 82% at 4×, 10×, 20×, and 40× magnification, respectively. Additionally, a comparative analysis with the existing state-of-the-art technique demonstrated the superiority of the proposed scheme.