Certain Investigations on Brain Tumor Localization, Segmentation and Classification Approaches with Research Recommendations for Refinement

Abstract

Brain tumor diagnosis, a gradual process indulges several techniques; aid the classification of brain tumor. The diagnosis procedure comprises, pre-processing, localization, feature extraction, segmentation and classification. Deep Learning (DL) algorithms support every diagnosis and classification process. Depending on the dataset used, the authors decide the Machine Learning (ML) algorithms either separately or fuse the algorithms and procedures to attain foolproof classification. The paper’s objective is to throw light on the procedures adopted in brain tumor detection and classification processes. The paper focuses on the conservative and contemporary approaches of the past two decades namely, (a) Threshold-based approaches, (b) Active-Contour Model-based approaches, (c) Bounding Box-based approaches, (d) Clustering-based approaches, (e) Genetic Algorithm-based Clustering approaches, (f) Texture-based Segmentation approaches, (g) Optimization-based approaches, (h) Phase Stretch Transform-based approaches and Hybridized-conventional approaches for optimum performance. Apart from the procedures of the prevailing algorithms, the performances of those methods were discussed in a precise manner, such as the dataset adopted, suitable ML models with its architecture and distinct performance metrics along with their significance and pitfalls. To conclude, the findings of the existing methods provide valuable insights for researchers in terms of research recommendations and opportunities for refinement, specifically in relation to brain tumor processing stages.