Home // eTELEMED 2025, The Seventeenth International Conference on eHealth, Telemedicine, and Social Medicine // View article

Zero-Shot Super-Resolution for Low-Dose CBCT Images Using Lightweight StereoMamba

Authors:
Simin Mirzaei
Zhenchao Ma
Hamid Reza Tohidypour
Panos Nasiopoulos

Keywords: Cone-Beam Computed Tomography; spatial resolution; StereoMamba; lightweight models; zero-shot learning.

Abstract:
Cone-Beam Computed Tomography (CBCT) is a crucial imaging tool in medical diagnostics, but low-dose scans—necessary for minimizing patient radiation exposure—often suffer from degraded spatial resolution. Enhancing the visual quality of these scans is essential for accurate diagnosis and treatment planning. This paper introduces two primary contributions to address this challenge. First, we systematically evaluate the effectiveness of state-of-the-art super-resolution techniques on low-dose CBCT images. Due to the scarcity of real CBCT datasets, which are limited by radiation exposure constraints, we explore the potential of pre-trained stereo super-resolution models originally developed for RGB images. Unlike traditional CBCT datasets that rely on artificially synthesized image pairs, we employ a zero-shot approach to assess the adaptability of these pre-trained models to CBCT imaging. Second, our analysis reveals that existing deep-learning-based super-resolution networks struggle to generalize effectively to CBCT data. To address this, we develop a lightweight adaptation of StereoMamba, a model optimized for natural images, and tailor it to the structural characteristics of CBCT scans. Without requiring additional training, our optimized network achieves the highest Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) scores among all tested models, significantly enhancing CBCT image quality. Our approach makes a significant contribution to improving the quality of low-dose CBCT imaging and provides a path forward for improving diagnostic accuracy and clinical outcomes without increasing radiation risk.

Pages: 58 to 62

Copyright: Copyright (c) IARIA, 2025

Publication date: May 18, 2025

Published in: conference

ISSN: 2308-4359

ISBN: 978-1-68558-270-8

Location: Nice, France

Dates: from May 18, 2025 to May 22, 2025