Machine Learning and Computer Vision Based Methods for Cancer Classification: A Systematic Review

Cancer remains a substantial worldwide health issue that requires careful and exact classification to plan treatment in its early stages. Classical methods of cancer diagnosis involve lab-based testing using biopsy, and imaging tests. Modern technologies may contribute effectively to speed up the diagnosis of cancer. Machine learning-based algorithms have been more prominent in cancer classification in recent years. These algorithms hold great promise in interpreting complex datasets and applying the learned knowledge to categorize unseen samples for cancer classification. In addition, many computer vision-based algorithms play a vital role in image pre-processing, segmentation, and feature extraction. This review article discusses nine major cancer types: carcinoma, sarcoma, neuroendocrine tumor, melanoma, lymphoma, germ cell tumor, leukemia, brain tumor, and multiple myeloma. We conducted a detailed survey of recent literature. We focused on systems that utilize clinical imaging modalities as input and preprocessing, segmentation, and feature extraction as intermediate stages with machine learning classifier as their concluding stage. We have examined the works that classify cancer as mentioned above types using machine learning algorithms. We have analyzed six prominent machine learning-based algorithms: Support vector machines, decision trees, random forest, Naïve Bayes, logistic regression, and K-nearest neighbors. This work also gives insights into various imaging modalities, such as Computed Tomography scan, histopathological images, dermoscopic images, and their utility in diagnosing cancer. In addition, the paper discusses the performance measures used for evaluating the efficiency of machine learning-based models, including accuracy, sensitivity, specificity, F1-score. We have reviewed various pre-processing and segmentation techniques suitable for clinical image-based cancer classification. This survey also discusses some significant challenges researchers face during cancer classification studies. The main objective of this systematic review is to provide researchers and medical experts with extensive knowledge of the present status of cancer classification with the aid of computer vision and machine learning-based systems. We intend to provide a foundation for enhanced cancer detection and therapy precision using these techniques. This effort eventually contributes to the progression of the field of cancer and the enhancement of patient predic