Application of Artificial Intelligence Machine Learning Models in Cancer Diagnosis

In the medical domain, artificial intelligence (AI) pertains to the utilization of machine learning models and algorithms for the analysis of intricate medical data. This paper demonstrates the application of learning models in cancer diagnosis through two innovative approaches: the Hybrid U-Net model and an improved MobileNet-V3 model, as well as a novel Multiple Instance Learning (MIL) model. The combination of these models has achieved significant improvements in cancer diagnosis accuracy, interpretability, and performance. The integration of the Hybrid U-Net with MobileNet-V3 effectively enhances image segmentation and feature extraction efficiency, particularly in analysis of medical imaging tasks such as skin cancer detection. Through hyperparameter optimization, these models perform exceptionally well with complex data, significantly improving performance metrics such as accuracy, PPV, NPV, and AUC. On the other hand, the MIL model, with its innovative attention mechanism, has shown considerable promise in improving the diagnosis of lung cancer by providing greater interpretability and aiding clinical understanding. However, despite the impressive performance of these models, the research also highlights limitations such as small sample sizes and reliance on existing segmentation methods, which affect the stability and broad applicability of the models.