Heuristic strategy-assisted transfer learning for cervical cancer detection with GAN-based data augmentation

Background Cervical cancer mostly occurs in women due to abnormal growth of the cell in cervix region. However, identifying abnormal cells is a significant issue in the computer-based diagnosis model. So, the continuous screening process is required for earlier diagnosis of cervical cancer. Utilizing deep learning techniques helps to differentiate cancer cells from normal cells. Problems The uneven distribution of data in the conventional model is challenging affects the classification results and hence reliable techniques are introduced for detecting cervical cancer. Thus, it slows down the training process and provides delayed treatment. Thus, a novel cervical cancer detection approach is implemented by transfer learning techniques that can be accomplished by various stages. Methods The different stages are (i) data collection, (ii) data augmentation, (iii) patch splitting, and (iv) cancer detection. The standard images are gathered from the publicly available resources. Next, the data augmentation is done through a Cycle Generative Adversarial Network (CycleGAN). Instead of processing the whole image, the CycleGAN model helps to process with smaller patches of images. Then, the splitting of image patches is further undergone in the cancer detection model using Transfer Learning (TL), which is the combined model of ResNet, VGG16, Xception, MobileNet, and DenseNet.Also, the fine-tuning of the weight and thresholds of classifiers is performed using the Improved Density Factor-based Honey Badger Algorithm (IDF-HBA). Finally, the outcomes are attained by taking the average of the obtained scores. Results The developed model achieves improved performance of 96.26% and 96.27% regarding accuracy and sensitivity. Significance This improved performance helps the developed model identifies the cancerous region and allows timely treatment for enhancing the survival rate of the women's. Henceforth, the earlier detection of cervical cancer helps to minimize treatment cost as well as improves the diagnosis performance.