Abstract
Bone metastases (BMs) represent the third most frequent site of distant spread after lung and hepatic deposits and are a major cause of morbidity owing to pain, pathological fracture, and neurological compromise. Early, reliable detection is hindered by the asymptomatic nature of many lesions, the limited sensitivity of conventional radiography, the confounding effects of age-related skeletal change, and significant inter-observer variability. Recent advances in deep learning (DL) offer an opportunity to overcome these limitations; however, clinically deployable solutions must achieve not only high accuracy on heterogeneous data but also transparent decision-making to foster trust among radiologists.
The A.I.B.M. project will develop, validate, and clinically integrate an interpretable DL framework capable of detecting and segmenting bone metastases in volumetric CT and MRI studies of the vertebral column and pelvic girdle. A retrospective cohort from Pauls Stradiņš Clinical University Hospital will be curated, anonymised, and annotated in three planes under expert supervision. Following rigorous pre-processing, data augmentation, and voxel-level harmonisation, a 3D convolutional architecture will be trained and optimised on high-performance computing resources.
Research Aim
To develop, validate, and clinically integrate an interpretable deep-learning framework that automatically detects, segments, and characterises bone metastases in volumetric CT and MRI studies of the vertebral column thereby improving diagnostic accuracy, reducing time to diagnosis, and enhancing radiologist confidence through transparent, explainable AI outputs.