Enhancing Tumor Segmentation Accuracy in Medical Imaging Through Deep Model Fusion and Multi-Mask Aggregation

Accurate segmentation of tumors in medical imaging plays a crucial role in early diagnosis, precise treatment planning, and effective monitoring of cancer progression. Building on this importance, this study proposes a comparative evaluation of multiple segmentation mask aggregation strategies aimed at enhancing tumor detection performance using artificial intelligence techniques. Several mask fusion approaches, including individual masks, union, intersection, and majority voting, were applied to tumor datasets derived from CT and MRI-images. The segmentation performance was quantitatively assessed using two standard overlap-based metrics: the Dice Coefficient and the Jaccard Index. Experimental results demonstrate that aggregated masks, particularly those derived from intersection and majority voting strategies, yield more consistent and accurate delineation of tumor regions across diverse patient cases. The best-performing methods achieved mean Dice scores above 60% and Jaccard indices exceeding 50%, highlighting a strong overlap with expert-annotated ground truth and conforming the effectiveness of the proposed approaches. In contrast, some individual masks exhibited lower stability and precision, showing higher performance variance across patients. These findings underscore the effectiveness of multi-mask fusion as a robust and reliable strategy for enhancing the consistency and accuracy of AI-based tumor segmentation systems. Overall, this work lays the groundwork for integrating ensemble-based segmentation outputs into clinical decision-support tools, thereby reducing diagnostic uncertainty and contributing to more efficient and reliable radiology workflows.