Brain tumor diagnosis techniques key achievements, lessons learned, and a new CNN architecture
Abd-Ellah, Mahmoud Khaled Bayoumi, Esraa Salah Ashraf, Shady Safy, Mohammed Salama, Gerges
Abd-Ellah, Mahmoud Khaled
Bayoumi, Esraa Salah
Ashraf, Shady
Safy, Mohammed
Salama, Gerges
Type
Supervisor
Date
2025-09-13
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Background A brain tumor is an abnormal tissue growth in the skull that can damage healthy brain areas by exert‑
ing pressure. While early detection is vital for prevention, accurate diagnosis with computer-aided design (CAD)
systems remains challenging due to variations in tumor shape and location.
Aim This paper provided a structured literature survey (SLS) of various machine learning (ML) and deep learning (DL)
techniques that were utilized in detection, classification, segmentation, and fusion-based diagnosis involving multiple
diagnostic systems and a newly designed convolution neural network (CNN) architecture.
Method The SLS was based on reliable papers in the Web of Science (WoS) database and was organized into three
phases. The first evaluated recent review papers, identified the number of methodological studies in each, focused
on authenticated publications, and analyzed their diagnostic approaches, ending with a critical assessment
of the reviews. The second examined recent methodological works in brain tumor diagnosis that were not covered
in those reviews, assessing each by its performance metrics. Across these phases, 320 authenticated studies were
analyzed. The final phase introduced the detecting and classifying brain tumors (DCBT) system.
Results This system combined transferred EfficientNet-B0 (TR_EffNetB0) with a newly developed dual-path CNN
architecture, attaining an accuracy of 98.5%.
Conclusion The SLS concluded with intuitive key achievements and lessons learned, which made future research
easier