Comparing Convolutional Neural Networks and Segmentation-Based Algorithms for Early Detection of Cancers Using MRI

Presenter: Dat Huu Nguyen

Faculty Sponsor: Reena Randhir

School: Springfield Technical Community College

Research Area: Biology

Session: Poster Session 2, 11:30 AM - 12:15 PM, Auditorium, A24

ABSTRACT

Time is critical to beating cancer, and early detection can provide patients with a larger window for treatment. Artificial Intelligence (AI) has been integrated into the medical imaging to assist radiologist with analyzing the Magnetic Resonance Image (MRI) scans. This study compares the effectiveness and accuracy of two different AI techniques, Convolutional Neural Networks (CNNs) and segmentation-based Algorithms, in early detection of brain tumors using published research in PubMed to evaluate performance differences and clinical feasibility of these approaches.
Both approaches utilized Magnetic Resonance Imaging (MRI) scans from publicly available repositories and hospital archives. Published research shows that segmentation-based algorithms can achieve overall performance levels around 90% while requiring fewer computational resources and smaller datasets, making them a more practical solution for hospitals with limited resources. Studies of CNN-based methods report accuracy levels around or above 90–95% depending on the dataset and task. CNNs generally achieve higher accuracy than traditional segmentation approaches; however, they require large datasets and powerful hardware.
CNNs and segmentation-based algorithms are both effective AI techniques that can improve the accuracy of brain tumors early detection and provide near expert level of accuracy and time. Deciding which techniques to deploy will depend on the primary objectives and resources available. Earlier and more accurate detection of brain tumors using AI-assisted imaging will help reduce diagnostic delays and improve treatment planning and patient outcomes. Research also indicates the need for explainable and clinically interpretable AI models to improve physician trust and support integration into routine clinical workflows.

RELATED ABSTRACTS