Objective:

To introduce a gradient-guided hybrid inpainting framework that integrates OpenCV Telea and LaMa, leveraging their strengths for improved reconstruction of brain MRI and classification of Alzheimer's disease.

Key Findings:

• Achieved an 8% reduction in mean squared error (MSE) compared to LaMa and 30% compared to OpenCV.
• Maintained a structural similarity index (SSIM) of approximately 0.93 and peak signal-to-noise ratio (PSNR) of approximately 25.7 dB.
• A VGG16 classifier trained on clean images achieved 94.35% accuracy on hybrid-inpainted data, showing only a 1.69 percentage point drop from the baseline accuracy of 96.04%.

Interpretation:

The proposed hybrid inpainting framework effectively balances anatomical accuracy and semantic plausibility, significantly enhancing the quality of MRI data for Alzheimer's disease classification.

Limitations:

• The study may be limited by the specific dataset used and the generalizability of the findings to other imaging modalities or populations.
• The computational overhead, while modest, may still pose challenges for real-time applications, necessitating further optimization.

Conclusion:

The hybrid inpainting approach demonstrates improved reconstruction and classification performance, supporting its potential use in AI-driven neuroimaging pipelines.