A Detailed review of recent progress in the cancer detection, leveraging a machine learning and a deep learning models to enhance cancer diagnostic capabilities.

Abstract

Purpose- Numerous individuals succumb to various deadly illnesses, with cancer ranking among the deadliest. Cancer means unnatural and unrestricted tissue growth within the body, with the potential to metastasize to other organs. Therefore, early detection is crucial to administer timely and precise treatment. Manual diagnosis and diagnostic errors have been implicated in the loss of many lives, prompting extensive research into automatic and accurate cancer detection methods at the early stages.

Methods- This paper conducts a comparative analysis of two Artificial Intelligence models, that are Machine Learning (ML) and Deep Learning (DL) models for diagnosing various kinds of cancer and explores recent advancements in detection methods. The study focuses on four types of cancer - Brain, Lung, Skin, and Breast also evaluates their detection using ML/DL techniques. A comprehensive review was conducted, incorporating a total of 128 papers, comprising 55 ML-based and 73 DL-based cancer identification techniques. Only peer-reviewed literatures published within the past five years (2018–2023) were included for analysis. Parameters such as publication year, features used, best model used, dataset used, and best accuracy were considered. Machine Learning (ML) and Deep Learning-based techniques for cancer detection were reviewed separately, with accuracy serving as the primary performance evaluation parameter to ensure consistency while assessing classifier efficiency.

Results- Among the reviewed research papers, Deep Learning methods achieved the highest accuracy of 99%, whereas Machine Learning methods attained 98.85%. The least accuracies recorded for Deep Learning and Machine Learning methods were 71% and 74%, respectively. Notably, for skin cancer detection, there was a variation in accuracy of approximately 29% between the lofty and least performing models. Furthermore, the study presents key findings and challenges linked with each type of cancer detection using a Machine Learning and a Deep Learning approaches. A comparison study between the finest and worse performing models, along with overarching key discoveries and difficulties, is provided for upcoming research considerations. Despite the analysis primarily focusing on accuracy as the performance parameter and considering various other parameters, the results underscore substantial room for enhancement in classification parameter efficiency.

Conclusion- This survey paper concludes that both techniques, i.e. a Machine Learning and a Deep Learning techniques show potential in early cancer detection across various types. However, specific challenges have been identified that must be overcome for the widespread adoption of these techniques in clinical settings. The results mentioned offer valuable perception for upcoming future research ventures in cancer detection, highlighting the imperative for ongoing progress in a Machine Learning and a Deep Learning based approaches to increase the diagnostic accuracy parameter and ultimately contributing for saving lives.