Preprocessing and Postprocessing for Robust Bone Tumor Detection: A Siamese-Guided Weighted Augmentation Pipeline

Bone tumor detection in medical imaging poses a significant challenge due to the variability in tumor appearances and the need for precise and efficient analysis. Accurate detection is crucial for timely diagnosis and treatment; however, traditional methods often rely on multiple models for different tasks, resulting in increased complexity and computational overhead. In this paper, we propose a novel approach that addresses these issues by integrating preprocessing, a multitask backbone, and Siamese- driven post-processing into a cohesive framework for classification, segmentation, and localization. Our method utilizes YOLOv11n to focus on relevant image regions and YOLOv11-seg for simultaneous classification, localization, and segmentation. Then, we use a Siamese network to refine segmentation outputs. This unified design reduces the number of parameters and enhances performance, offering a promising solution for efficient and accurate bone tumor detection.