Multi-level thresholding segmentation for brain tumor detection using optimized deep learning approach

• Published in: Neural computing & applications Vol. 37; no. 23; pp. 19279 - 19302
• Main Authors: Ewees, Ahmed A., Ismail, Fatma H., Labeeb, Nada S., Gaheen, Marwa A.
• Format: Journal Article
• Language: English
• Published: London Springer London 01.08.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial Intelligence; Brain; Brain cancer; Brain research; Classification; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Deep learning; Effectiveness; Efficiency; Image Processing and Computer Vision; Image quality; Image segmentation; Machine learning; Medical imaging; Methods; Neural networks; Optimization algorithms; Original Article; Performance measurement; Probability and Statistics in Computer Science; Signal to noise ratio; Tumors; Brain tumor; Long short-term memory; Weighted mean of vectors algorithm; Multi-level thresholding segmentation
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-025-11398-w

Medical image segmentation and classification are essential for diagnosing brain tumors, which carry a high risk due to the intricate nature of the brain. In this study, we introduce a multi-level thresholding (MLT) approach to achieve high-quality segmentation. The proposed method utilizes deep learning techniques, employing the weighted means of vectors optimization (INFO) algorithm to determine MLT threshold values. Additionally, we fine-tune the hyperparameters of the long Short-term memory (LSTM) neural network, a powerful deep learning architecture, using the INFO algorithm. This LSTM network is employed for classifying the segmented images and detecting brain tumors. The effectiveness of the proposed deep learning approach is assessed on a Kaggle dataset comprising 253 MRI images, including 98 non-tumor and 155 tumor images. We employ performance metrics, including peak signal-to-noise ratio, structural similarity Index, sensitivity, specificity, and accuracy, to evaluate the quality of the deep learning-based segmentation and classification process. The proposed deep learning-based approach consistently outperforms other methods in terms of producing high-quality segmented tumor images and exhibits enhanced detection and classification performance. These findings highlight the effectiveness of the proposed approach in segmenting and classifying brain tumor images.