ADVANCED MULTI-STAGE PRE-PROCESSING PIPELINE FOR ROBUST THYROID NODULE US IMAGE ENHANCEMENT AND DIAGNOSTIC FEATURE EXTRACTION

Abstract

One of the most important challenges in ultrasound imaging is the diagnosis of thyroid nodules because of variability in images, ultrasound noise, and artifacts. This applied research proposes a blend hybrid noise reduction, machine-learning assisted normalization, and intelligent segmentation algorithms into a multi- stage ultrasound image processing pipeline. The proposed methodology outperformed traditional techniques and achieved an overall diagnostic accuracy of 96.2% with 89.4% sensitivity, 92.1% specificity, and Dice Similarity Coefficient (DSC) of 0.886. Furthermore, it has SSIM of 0.92, and average throughput processing time of 52.3 ms per image. The use of adaptive filtering combined with deep learning feature boundaries preservation techniques profoundly improves the quality of ultrasound images and the accuracy of its diagnosis across varying clinical datasets.