Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study evaluates the effectiveness of user-prompted priors, specifically multi-plane contour inputs, in enhancing semi-automated segmentation of cancer lesions in whole-body CT scans. By employing a 3-fold cross-validation and external testing on 3,865 lesions, the research demonstrates that complex spatial priors significantly improve segmentation accuracy, achieving a mean Dice score of 0.882 compared to 0.671 for the baseline model. These results underscore the potential of user-guided approaches to streamline the generation of high-quality ground-truth data for tumor volume assessment, which is crucial for better predicting patient outcomes.
Multi-plane user-prompted priors can boost cancer lesion segmentation accuracy by over 30%, revolutionizing how we assess tumor burden in clinical oncology.
In clinical oncology studies, metastatic cancer is commonly evaluated using "Response Evaluation Criteria in Solid Tumors" (RECIST), in which the diameter of up to five lesions is measured and followed over the course of treatment. However, RECIST shows limited correlation with overall survival. Total tumour volume (TTV) is a stronger predictor but typically relies on manual ground-truth segmentation of all lesions, which is time-consuming and requires expert domain knowledge. Semi-automated approaches leveraging user-prompted priors, such as bounding boxes and single-slice contours, as inputs to automated segmentation methods can facilitate the generation of ground-truth segmentations. This work investigates the impact of different user-prompted priors on semi-automated cancer lesion segmentation performance in whole-body computed tomography. Across 3-fold cross-validation and external testing, more complex spatial priors consistently improved performance, with contour priors from three orthogonal planes (axial, coronal and sagittal) achieving the best results. On the external test (n=3865 lesions), this approach achieved a mean Dice score of 0.882, compared to a mean Dice score of 0.671 for the baseline model with no spatial prior. These findings suggest that the use of multi-plane orthogonal user-prompted priors can improve semi-automated tumour lesion segmentation and support efficient generation of high-quality volumetric ground-truth data.