M. Rahman

What I work on

Deep learning for computer vision — with a focus on medical image analysis, spectral/remote-sensing imagery, and trustworthy models.

Medical image analysis

My core line of work applies deep learning to medical imaging: my MSc thesis and UGC-funded project developed efficient semantic segmentation models for brain tumors in MRI, including an attention-refined U-Net with skip connections (ICCIT 2023) and an ensemble transfer-learning approach (BIM 2023). Related collaborations extended these ideas to knee injury classification from MRI and award-winning skin lesion detection (Best Paper, ICISET 2024).

Remote sensing & spectral imaging

Starting from my BSc thesis on hyperspectral image classification with factor analysis and CNNs (Springer, 2022), I collaborate on satellite-imagery problems — gradient-based channel selection for bushfire classification (ECCE 2025) and cross-sensor domain adaptation between Landsat and Sentinel-2, currently under journal review.

Multimodal & language applications

I also explore multimodal learning beyond vision: Bengali music genre classification combining 1D CNNs and ResNet-50 (ICCIT 2024), Bengali sentence categorization, and NLP/ML system evaluation as Focal Point for the national Bangla Machine Translator testing effort.

Where I'm headed

I am working toward trustworthy, data-efficient vision models for clinical use — models that quantify uncertainty, adapt across imaging domains, and remain reliable where labeled data is scarce.