Ovarian cancer identification using transfer learning
Rishabh Dhenkawat, Samridhi Singh, Nagendra Singh · 2023
Ovarian cancer as it is more often called, is women&s;s fifth largest cause of cancer-related mortality. As of late, deep learning has been proven superior to traditional approaches for predicting OC stages and subtypes. Despite this, most state-of-the-art deep learning models only use data from a single modality, which might lead to subpar performance due to an inadequate portrayal of crucial OC features. In addition, the quantity of computational resources needed for training and deploying these deep learning models is greatly increased due to the absence of optimization of the model architecture. In this research, we create a hybrid evolutionary deep learning model with seven unique architectures and evaluate their performance. In the end, a multi-model ensemble architecture is introduced. The deep feature extraction network that we built up independently on vgg16, vgg19, resnet, and ensembled of all of these architectures served as the basis for generating each modality&s;s different states and forms. After analysis and comparison, the final ensemble output is produced. Generalization Stacking As a meta learner, multi-layer perceptron is employed due to its rapid training time and superior accuracy relative to alternative methods.