U-NET, a convolutional neural network architecture, demonstrates exceptional performance in brain tumour segmentation, achieving 98.56% accuracy and 99% F-score in MRI image analysis. Brain health is crucial for overall well-being, with magnetic resonance imaging (MRI) playing a vital role in diagnosing neurological conditions. As these scans generate vast datasets, deep learning techniques have emerged as powerful tools for image processing and classification. The study focused on classifying brain tumours, including glioma, meningioma, and pituitary tumours, using the U-Net architecture and exploring the effectiveness of other convolutional neural networks such as Inception-V3, EfficientNetB4, and VGG19, enhanced through transfer learning.
The research employed various evaluation metrics to assess model performance, including F-score, recall, precision, and accuracy. The U-Net segmentation architecture emerged as the top performer, achieving an impressive accuracy of 98.56%, an F-score of 99%, an area under the curve of 99.8%, and both recall and precision rates of 99%. Notably, U-Net demonstrated robust performance across diverse clinical scenarios, achieving 96.01% accuracy in cross-dataset validation with an external cohort.
These findings underscore the potential of U-Net and transfer learning techniques in enhancing diagnostic accuracy and informing clinical decision-making in neuroimaging. The high performance of U-Net in brain tumour segmentation holds promise for early detection and treatment planning, potentially improving patient care and outcomes. As the field of medical imaging continues to evolve, the integration of advanced deep learning models like U-Net into clinical practice could revolutionise the diagnosis and management of brain tumours. Future research should focus on validating these results in larger, diverse patient populations and exploring the potential of these technologies in real-world clinical settings to further improve neurological care.
Jenna Lorge, EMJ
Reference
Ilani MA et al. T1-weighted MRI-based brain tumor classification using hybrid deep learning models. Sci Rep. 2025;DOI:10.1038/s41598-025-92020-w.
