Evolutionary Advances in Machine Learning and Deep Learning for Cancer Detection and Classification

Advances in artificial intelligence (AI) have transformed the landscape of medical diagnostics, offering unprecedented opportunities for early disease detection and classification. Among critical health challenges, cancer continues to demand reliable and timely diagnostic solutions that overcome the limitations of conventional methods such as subjectivity, labor intensity, and variable reproducibility. Over the past decade, AI methodologies in oncology have undergone a remarkable evolution. Initially, conventional machine learning (ML) models were widely used. These approaches relied heavily on handcrafted features designed by domain experts. Subsequently, deep learning (DL) architectures emerged, enabling automatic feature extraction directly from raw medical data such as images and gene-expression profiles. More recently, transformer-based and multimodal frameworks have been introduced. These advanced models are capable of integrating multiple data types—imaging, genomic, and clinical information—to provide a more holistic understanding of cancer biology and patient outcomes. Quantitative comparisons reveal clear performance advancements: ML techniques such as Support Vector Machines (SVM) and Random Forests achieved accuracies of approximately 82–84% with AUCs near 0.85; DL architectures like CNNs and Efficient Net improved outcomes to around 91–93% accuracy and ~0.94 AUC; transformer-based models further enhanced results to ~95% accuracy and ~0.96 AUC. The most promising performance emerged from multimodal fusion frameworks, which reached ~97% accuracy and ~0.98 AUC, demonstrating the power of holistic data integration in capturing complex cancer signatures.