OCEAN - Ovarian Cancer subtypE clAssification and outlier detectioN using DenseNet121

In our exploration, we tackle the significant hurdle of diagnosing ovarian cancer, a formidable challenge within the field of oncology. This cancer type often remains undetected until its advanced stages due to inconspicuous symptoms and the absence of reliable early detection methods. To combat this issue, we propose a method utilizing DenseNet121, a deep learning approach, for the multi-classification of ovarian cancer sub-types based on histopathology images. We leverage the UBC Ovarian Cancer data-set obtained from Kaggle, encompassing whole slide images and tissue microarrays with various dimensions. Among the sub-types covered in this dataset are clear cell (CC), endometrioid (EC), high-grade serous (HGSC), low-grade serous (LGSC) and mucinous (MC). To address the challenge posed by large image sizes up to 100,000 x 50,000 pixels and the diverse morphological characteristics present in the images, we have devised comprehensive data augmentation techniques. Our DenseNet121 model demonstrates robust generalization capabilities by effectively learning features from this varied data-set. A thorough experiment with challenging cases of ovarian cancer sub-types resulted in 99.6% accuracy and 100% F1-score in the multi class classification. Ovarian cancer detection systems based on this approach may be developed to detect the disease early and accurately.