Neural network method boosts medical image compression quality at high ratios

A frequency-adaptive neural compression method outperformed HEVC and JPEG2000 on abdominal CT. At high compression, it achieved peak PSNR of 50.76, SSIM of 0.9930, and the lowest RMSE (0.0029), with significantly higher subjective quality scores.

• The proposed FAINC method uses a frequency-domain gating mechanism to dynamically allocate image blocks to subnetworks of different capacities, improving reconstruction.
• Ablation studies showed the gating mechanism and dynamic parameter allocation contributed approximately 2.05 dB and 1.87 dB PSNR improvements, respectively.
• The study retrospectively analyzed abdominal CT data from 356 patients across the KiTS19 and AVT datasets.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.