Multi-Cancer Classification using InceptionV3 + Vision Transformer

This research proposes a two-architecture system for multi-cancer classification and diagnosis using the InceptionV3 and Vision Transformer (ViT) since CNN has known drawbacks in cancer diagnosis. The different cancers of interest in the proposed system are cervical, brain, kidney, lymphomas, lung, colon, oral, and breast cancer to improve on early detection and accurate diagnosis of the diseases. InceptionV3 is used in feature extraction which extracts fine details in medical images while the Vision Transformer analyzes such features to better classify relationships between images. This two-architecture strategy takes advantage of the proposed models and eliminates some inherent drawbacks of stand-alone CNN models such as VGGNet and EfficientNet. Empirical tests carried out on large real-life data sets show that the proposed model yields a higher accuracy, precision, and recall than previous solutions and benchmarks. The utility of the given hybrid system, in turn, is based on its ability to work with different types of multi-class classification problems and provide an accurate automated diagnostic solution. A potential for clinical uses, the outlined approach minimizes the role of human mistakes, optimizes cancer detection, and ultimately optimizes patient outcomes due to early and precise diagnosing.