Automatic intracranial abnormality detection and localization in head CT scans by learning from free-text reports

Deep learning has yielded promising results for medical image diagnosis but relies heavily on manual annotations, which are expensive to acquire. We present Cross-DL, a cross-modality framework intracranial abnormality detection and localization in head computed tomography (CT) scans by from free-text imaging reports. Cross-DL discretizer that automatically extracts discrete labels of types locations reports, utilized train an analyzer dynamic multi-instance approach. Benefiting the low annotation cost consequent large-scale training set 28,472 CT scans, achieves accurate performance, with average area under receiver operating characteristic curve (AUROC) 0.956 (95% confidence interval: 0.952–0.959) detecting 4 17 regions while accurately localizing abnormalities at voxel level. An hemorrhage classification experiment external dataset CQ500 AUROC 0.928 (0.905–0.951). The model can also help review prioritization.