Machine learning for brain cancer diagnosis: Systematic literature review (#1116)

Authors

Ari Coaquira, Roly Antony

Martinez Saldaña, Sandra Carolina

Sanchez Anastacio, Katherine Rosemary

Alarcon Vasquez, Felipe

Abstract

The diagnosis of brain tumors remains one of the greatest challenges in modern medicine due to the complexity and subjectivity of medical imaging interpretation methods. The purpose of this review is to evaluate the accuracy of machine learning techniques in diagnosing brain tumors and compare them with the manual interpretation of medical images performed by radiologists. Thirty original articles from Scopus and IEEE Xplore were selected, all of which were no older than five years. The review followed the PICO and PRISMA methodologies for the collection and analysis of the studies. The results show that machine learning techniques, particularly convolutional neural networks (CNN), achieve high accuracy, in some cases exceeding 90%, outperforming traditional methods such as Support Vector Machines (SVM) and K-Nearest Neighbors (KNN). Hybrid models like MobileNetV2 combined with VGG19 have demonstrated an accuracy of 92-95%, with advantages in processing speed. Additionally, it was identified that the availability and quality of labeled data significantly influence the accuracy of the models, with optimal performance in studies with large volumes of data. Among the most commonly detected tumors using these techniques are gliomas, meningiomas, and glioblastomas, although results have also been achieved in less frequent tumors such as ependymomas and medulloblastomas. In conclusion, machine learning techniques, particularly CNNs, show high potential for improving the accuracy of brain tumor diagnosis. Further research is needed to optimize their application and integration into clinical practice.