Performance of Quality Metrics for Compressed Medical Images Through Mean Opinion Score Prediction

This paper examines the performance of two objective quality assessment metrics; peak signal-to-noise ratio (PSNR) and Structural SIMilarity (SSIM) index for compressed medical images through subjective mean opinion score (MOS) prediction. MOS prediction models have been developed by establishing mathematical relation between the theoretically computed objective (PSNR and SSIM) and subjective Mean Opinion Score (MOS) quality parameters. Based on the developed prediction models, MOS prediction values have been generated for varying PSNR and SSIM values for compressed MRI and ultrasound images. It is observed that for same value of PSNR/SSIM, MOS values are different depending on the type of compression technique used. It is found that SPIHT scheme gives higher predicted MOS values as compared to JPEG and JPEG2000 schemes at lower PSNR (<= 38 dB) for considered MR and ultrasound images. SPIHT scheme also gives higher predicted MOS values at lower SSIM (<= 0.75) values for MR images but for ultrasound images JPEG2000 gives better predicted MOS values at SSIM (<= 0.90). This paper also provides information about correlation coefficient (CC) between peak signal to-noise ratio (PSNR)/Structural SIMilarity (SSIM) index and experimental subjective quality metrics. It is observed that PSNR gives better correlation with MOS values for all compression schemes.