Patch-Wise Local Principal Component Analysis-based Medical Image Denoising: A Method Noise Approach

Gaussian noise is often added during the acquisition or transmission of medical images, which can blur important organs and reduce diagnostic accuracy. To address this issue, a hybrid denoising model that combines BayesShrink thresholding in the wavelet domain with Patch-Wise Local Principal Component Analysis (PLPCA) is proposed.

The framework initially utilises the concept of local patch redundancy by applying PLCA to effectively suppress noise while preserving finer details. The remaining noise is again decomposed using the wavelet transform, with adaptive BayesShrink thresholding applied to refine the coefficients. Subsequently, the reconstructed signal is obtained. The resultant denoised images are then combined to obtain the final enhanced image. HRCTs and MRI datasets were corrupted by Gaussian noise at σ = 10-40 and were validated.

Quantitative analysis based on PSNR, entropy, BRISQUE, NIQE, and PIQE confirmed the consistent high impact of the proposed method across traditional, hybrid, and deep learning baselines. It is worth noting that the approach achieved PSNR values of 31.15 dB at HRCT and 33.00 dB at MRI when σ = 10, as well as at high noise levels (23.95 dB and 22.64 dB, respectively).

The proposed method consistently excels over traditional, hybrid, and deep learning-based strategies, as shown by quantitative assessments using PSNR, entropy, BRISQUE, NIQE, and PIQE.

Conclusion

The suggested framework has the strength of suppressing Gaussian noise while retaining anatomical detail. Thus, it is an encouraging option for enhancing the quality of medical image diagnosis in clinical practice.