Cervical Cancer Screening on Multi-class Imbalanced Cervigram Dataset using Transfer Learning

Image classification from a multi-class imbalanced dataset is challenging because it is difficult to detect all the minority classes present in the datasets. In this paper, the authors have extended their recently introduced work on a novel deep learning neural network for binary-class imbalanced datasets, called VGGIN-Net, by applying it to classify different grades of cervical cancer from a multi-class imbalanced dataset having a small number of samples. The experimental results prove that the proposed approach along with the data augmentation and rejection resampling is effective for multi-class imbalanced datasets. As analyzed through extensive experiments on a benchmark cervical screening dataset, the proposed approach in comparison to the other state-of-the-art approaches is vividly proven to be a more efficient method.