The classification of the bladder cancer based on Vision Transformers (ViT)

Bladder cancer is a prevalent malignancy with diverse subtypes, including invasive and non-invasive tissue. Accurate classification of these subtypes is crucial for personalized treatment and prognosis. In this paper, we present a comprehensive study on the classification of bladder cancer into into three classes, two of them are the malignant set as non invasive type and invasive type and one set is the normal bladder mucosa to be used as stander measurement for computer deep learning. We utilized a dataset containing histopathological images of bladder tissue samples, split into a training set (70%), a validation set (15%), and a test set (15%). Four different deep-learning architectures were evaluated for their performance in classifying bladder cancer, EfficientNetB2, InceptionResNetV2, InceptionV3, and ResNet50V2. Additionally, we explored the potential of Vision Transformers with two different configurations, ViT_B32 and ViT_B16, for this classification task. Our experimental results revealed significant variations in the models’ accuracies for classifying bladder cancer. The highest accuracy was achieved using the InceptionResNetV2 model, with an impressive accuracy of 98.73%. Vision Transformers also showed promising results, with ViT_B32 achieving an accuracy of 99.49%, and ViT_B16 achieving an accuracy of 99.23%. EfficientNetB2 and ResNet50V2 also exhibited competitive performances, achieving accuracies of 95.43% and 93%, respectively. In conclusion, our study demonstrates that deep learning models, particularly Vision Transformers (ViT_B32 and ViT_B16), can effectively classify bladder cancer into its three classes with high accuracy. These findings have potential implications for aiding clinical decision-making and improving patient outcomes in the field of oncology.