MRI Image Enhancement Using the MWACNN Method and Comparison with U-Net
For my master's thesis, I investigated deep-learning approaches to reducing motion artifacts in MRI scans. I trained and evaluated four architectures under comparable conditions using the KMAR-50K dataset.
Key Research Findings
U-Net achieved the strongest overall reconstruction quality. Channel-attention MWACNN approached U-Net on some measures while using substantially fewer trainable parameters.
Compared models: U-Net, WACNN, MWACNN, and channel-attention MWACNN. Experimental research findings; not a clinical validation.