Toward Better Optimization of Low-Dose CT Enhancement: A Critical Analysis of Loss Functions and Image Quality Assessment Metrics

TL;DR

The paper investigates how loss functions used to train LDCT image enhancement relate to image quality as measured by FR and NR IQA metrics. By evaluating pixel-based (L1, MSE, Charbonnier) and feature-based (VGG) losses within a representative LDCT model on the Mayo2016 dataset, it reveals limited and inconsistent alignment between loss optimization and perceptual/diagnostic quality. Although VGG-based perceptual losses show some coherence with FR perceptual metrics, most IQA metrics behave erratically during training, indicating that current metrics and losses do not robustly guide enhancement toward clinically meaningful quality. The work highlights the need for perceptually aligned loss designs and more reliable IQA metrics to better drive LDCT enhancement toward meaningful diagnostic improvements.

Abstract

Low-dose CT (LDCT) imaging is widely used to reduce radiation exposure to mitigate high exposure side effects, but often suffers from noise and artifacts that affect diagnostic accuracy. To tackle this issue, deep learning models have been developed to enhance LDCT images. Various loss functions have been employed, including classical approaches such as Mean Square Error and adversarial losses, as well as customized loss functions(LFs) designed for specific architectures. Although these models achieve remarkable performance in terms of PSNR and SSIM, these metrics are limited in their ability to reflect perceptual quality, especially for medical images. In this paper, we focus on one of the most critical elements of DL-based architectures, namely the loss function. We conduct an objective analysis of the relevance of different loss functions for LDCT image quality enhancement and their consistency with image quality metrics. Our findings reveal inconsistencies between LFs and quality metrics, and highlight the need of consideration of image quality metrics when developing a new loss function for image quality enhancement.

Figures (4)

• Figure 1: Evaluation of Loss Functions in DenoMamba: Training Loss Curves for Charbonnier, L1, MSE, and VGG loss
• Figure 2: Reconstructed LDCT images using different loss functions. (a) LDCT input, (b-e) reconstructions using Charbonnier, L1, MSE, and VGG-based models, respectively. (f) HDCT reference. The display window is set to [-160, 240] HU.
• Figure 3: Training Progress: Full-Reference QA Metrics (PSNR, SSIM, VIF, LPIPS, ST-LPIPS, DISTS)
• Figure 4: Training Progress: No-Reference IQA Metrics