M-Net: An encoder-decoder architecture for medical image analysis using ensemble learning

• Published in: Results in engineering 2023-03, Vol.17, p.100927, Article 100927
• Main Authors: S, Sreelakshmi, G, Malu, Sherly, Elizabeth, Mathew, Robert
• Format: Article
• Language: English
• Subjects: Alzheimer's disease; Classification; Deep encoder-decoder network; Ensemble learning; Segmentation; sMRI
• ISSN: 2590-1230; 2590-1230
• DOI: 10.1016/j.rineng.2023.100927

Only a few of the many subfields of biomedical science study include biomedical engineering, biomedical signal processing, gene analysis, and biomedical image processing. For the investigation and diagnosis of diseases, classification, detection, and recognition have tremendous importance. This work presents a fully automated deep-ensemble architecture, M-Net, for pixel-level semantic segmentation and classification of medical images. The performance of M-Net is evaluated by implementing it on the brain structural Magnetic Resonance Imaging (sMRI) for diagnosing Alzheimer's disease from various sources of datasets. The M-Net system successfully segmented the hippocampus region, vulnerable to damage at the early stage of AD, from the brain sMRI data. The obtained overall accuracy of 99% shows that the proposed deep learning technique is superior to the existing deep semantic segmentation techniques and can reduce the diagnostic time of radiologists. •Introduces an efficient encoder-decoder model for medical image analysis.•The model is capable of segmenting brain regions from multiple modalities.•The evaluation has been done on real-world MRI datasets for Alzheimer's disease classification.•The proposed method is found to be outperforming chosen baselines.