Parameter-Efficient CT Reconstruction via Deep Graph Laplacian Regularization
2026-05-25 • Artificial Intelligence
Artificial IntelligenceComputer Vision and Pattern RecognitionMachine LearningSymbolic Computation
AI summaryⓘ
The authors looked for a way to improve low-dose CT scan images using much fewer computer resources than usual. They combined a special math tool called graph-based regularization with small neural networks in a new method called Deep GLR. This approach improved image quality a lot compared to a basic method, even though it used about 5 times fewer parameters and was trained on much less data. Their results suggest that graph-based regularization helps capture important image details efficiently. However, some gap in performance remains compared to the most advanced methods that use much bigger models and data.
Low-dose computed tomography (LDCT)Graph-based regularizationDeep Graph Laplacian Regularization (Deep GLR)Proximal Forward-Backward SplittingConvolutional Neural Networks (CNN)Peak Signal-to-Noise Ratio (PSNR)Filtered backprojectionParameter efficiencyData efficiencyImage reconstruction
Authors
Veera Varuni Radhakrishnan, Chinthaka Dinesh, Qurat-ul-Ain Azim
Abstract
Low-dose computed tomography (LDCT) reconstruction faces a critical tradeoff between reconstruction quality and resource requirements. While recent deep learning methods achieve state-of-the-art performance, they typically rely on over 500,000 parameters trained on large-scale datasets exceeding 35,000 scans. This work investigates whether graph-based regularization can provide meaningful noise reduction under strict resource constraints. We propose Deep Graph Laplacian Regularization (Deep GLR), integrating quadratic graph regularization into a Proximal Forward-Backward Splitting optimization framework with three lightweight CNN modules. Evaluated on the LoDoPaB-CT benchmark, Deep GLR achieves 30.70 dB PSNR, representing a 6.33 dB improvement over filtered backprojection, while using only 91,848 parameters trained on 1000 samples (2.8\% of standard training set). Compared to benchmark methods, this represents 5.8 times better parameter efficiency and 30 times better data efficiency per dB improvement. The learned graph bandwidth parameter ($ε$=1.25) converges to interpretable values, suggesting the method captures meaningful image priors rather than overfitting. While a 13 dB gap remains versus state-of-the-art methods, results demonstrate that graph-based regularization provides a favorable efficiency-quality tradeoff for resource-constrained medical imaging scenarios.