Medical Imaging 2.0: A New Look at Physical Measurements from Dynamic Biological Systems of in vivo Tissues for Early Cancer Detection

Le résumé fourni par la source

Expert interpretation of medical images for lesion diagnosis may not always align with the underlying in vivo tissue pathology and thus cannot be considered a definitive truth for malignancy or benignity. This discrepancy poses a major barrier to effective low dose computed tomography (LdCT) screening for early cancers, as follow-up evaluations can be costly and risky. We hypothesize that the barrier stems from the static nature of conventional image representations, which fail to capture the dynamic biological behavior of tissues, where early pathological changes are typically expressed. Current imaging technologies lack the spatial, temporal, and spectral (energy) resolution necessary to address the limitation. To test the hypothesis, we model the physical measurements as dynamic responses of biological systems to Xray flux across a broad energy spectrum. The interaction generates intrinsic image contrasts for each tissue type (e.g., muscle, fat, lung, bone) and energy level, thereby encoding biological information in the multispectral contrasts. For lesion diagnosis task, we reconstruct multispectral image contrasts across the lesion space at individual energy levels. Among various biologically relevant markers of pathological change, we focus on tissue elasticity and develop a model to compute elastic characteristic features (ECFs). Correspondingly we design an adaptive classifier for the multispectral ECFs. Evaluation was conducted using five datasets: two sets of pulmonary nodules and three sets of colorectal polyps. All lesions have pathology reports as ground truth. The presented method achieved area under receiver operating characteristic curve (AUC) values of 0.89–0.99, representing a significant improvement over state-of-the-art machine learning models, which report AUCs of 0.60–0.95. The high AUCs observed suggest strong potential for overcoming the current barrier in LdCT screening of early lung and colorectal cancers.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.