| |
 |
UNIVERSITY OF BUCHAREST
FACULTY OF PHYSICS Guest
2026-06-11 23:58 |
 |
|
|
|
Conference: Bucharest University Faculty of Physics 2026 Meeting
Section: Biophysics; Medical Physics
Title:
Physically-Constrained SRCNN for CT Image Super-Resolution
Authors:
Ilinca-Alexandra VOICAN (1), Leonard GEBAC (1), Vasile BERCU (1)
Affiliation:
1) Faculty of Physics, University of Bucharest, 077125 Magurele, Ilfov, Romania
E-mail
ilinca-alexandra.voican@s.unibuc.ro
Keywords:
Artificial Intelligence, Physics-constrained deep learning, SRCNN, CT imaging
Abstract:
Computed Tomography (CT) imaging plays a critical role in modern medical diagnosis and treatment planning. However, there is a trade-off between image quality and radiation dose reduction requirements. Due to acquisition constraints and hardware limitations, CT images may exhibit reduced spatial resolution and increased artifacts, making post-reconstruction enhancement methods essential for improving image quality and anatomical visibility. In this work, it was developed a deep learning model for increasing the resolution of CT images while incorporating physically-motivated mass conservation constraints. For simplicity and interpretability, the selected architecture is the Super-Resolution Convolutional Neural Network (SRCNN). The model was trained for this specific purpose on a relatively large and homogeneous dataset of CT images obtained from The Cancer Imaging Archive (TCIA), following an extensive image processing stage designed to preserve physical consistency. After developing the SRCNN model with global and local mass conservation constraints, introduced empirically into the loss function, the results indicated that the proposed local physical constraints encourage the reconstructed super-resolution images to preserve mass consistency while also having a favorable impact on reconstruction quality. As such, after testing the performance of the model, we obtained the following values of the evaluation metrics: PSNR=46.19 dB, SSIM=0.9843.
References:
-
Acknowledgement:
-
|
|
|
|