Optimized Enhancement of Pelvic CT Scan Images using Image Processing Techniques

Abstract

Computed Tomography (CT) imaging plays a crucial role in diagnosing pelvic disorders such as cancers, vascular abnormalities, and traumatic injuries. However, reducing radiation dose in pelvic CT scans—a key priority under the ALARA (As Low As Reasonably Achievable) principle—often leads to in creased noise and reduced contrast, impairing diagnostic accuracy. Addressing this trade-off, the present research aims to minimize radiation exposure while maintaining diagnostic-quality imaging through a hy brid image enhancement framework combining Non-Local Means (NLM) filtering for noise suppression and wavelet-based transform techniques for contrast enhancement. A dataset of 140 anonymized low-dose pelvic CT images from Thellipalai Base Hospital, Jaffna, was used. The proposed two-stage approach first applies NLM denoising to reduce random noise while preserving fine structures, followed by wavelet decomposition and thresholding to selectively enhance high-frequency details. A grid search across NLM smoothness pa rameters (5–30) and wavelet thresholds (0.01–0.09) identified the optimal configuration (h = 5, threshold = 0.01). Quantitative analysis using Peak Signal-to-Noise Ratio (PSNR) and Signal-to-Noise Ratio (SNR) showed marked improvement over standalone enhancement methods, achieving mean PSNR above 44 dB and substantially enhanced SNR values. These improvements indicate clearer structural delineation and reduced visual fatigue, as confirmed by radiologist feedback. Compared with conventional single-stage methods, the hybrid approach demonstrated superior balance between noise reduction and anatomical detail preservation, directly supporting accurate diagnoses without additional radiation burden. This research highlights how optimized hybrid enhancement can bridge the gap between dose reduction and image interpretability, offer ing a scalable, low-cost solution for clinical settings—especially in low-resource environments. Future work will integrate machine learning– based adaptive parameter tuning to further automate and personalize the enhancement process for diverse diagnostic needs.