Prediction of brain age from routine T2-weighted spin-echo brain magnetic resonance images with a deep convolutional neural network

Deep learning based automatic detection algorithm for acute intracranial haemorrhage: a pivotal randomized clinical trial

“…With respect to the AUROC values, the performance of the standalone AI algorithm in the external validation study (0.992 and 0.977 in patient-and slice-wise analyses, respectively) and reader assessment study (0.9874 and 0.9671 in patient-and slice-wise analyses, respectively) were comparable with the performance of the neuroradiologist subgroup without AI assistance (0.9764 and 0.9691 in patient-and slice-wise analyses, respectively). These diagnostic accuracies were higher than those reported by the majority of previous studies 7,8,10,11,13,15 and were comparable with the results achieved in a previous study (AUROC = 0.991), which reported that AI standalone performance was comparable with that of highly trained experts 13 . Furthermore, in the present study, the high sensitivity of 95.89% and specificity of 95.33% achieved by our approach at a cut-off level of 39.84% in the patient-wise analysis was higher than those achieved by reviewers without AI assistance (94.37% and 95.04%, respectively).…”

“…Despite the clinical relevance of diagnosing AIH using brain CT scans—false negatives may delay correct diagnosis, which can cause devastating consequences, whereas false positives will lead to unnecessary examinations—prompt and accurate assessment of AIH using brain CT scans remains a challenge for physicians. In addition, the high volumes of imaging data that require assessment place a significant burden on radiologists who need to maintain diagnostic accuracy and efficiency 7 , 8 .…”

“…Nevertheless, non-contrast brain CT scans are the most widely used first-line diagnostic approach for identifying AIH due to the several disadvantages of MRI scans, including their limited availability, long image acquisition times, high cost, and issues with patient tolerance 5,6 Despite the clinical relevance of diagnosing AIH using brain CT scans-false negatives may delay correct diagnosis, which can cause devastating consequences, whereas false positives will lead to unnecessary examinations-prompt and accurate assessment of AIH using brain CT scans remains a challenge for physicians. In addition, the high volumes of imaging data that require assessment place a significant burden on radiologists who need to maintain diagnostic accuracy and efficiency 7,8 .…”

“…These innovations may increase diagnostic accuracy, enable prompt diagnosis and improved management of various conditions, and facilitate new biological insights. Various AI algorithms for AIH diagnosis have been developed and shown promising results in the detection, classification, quantification, and prediction of AIH using brain CT scans 7,8,[10][11][12][13][14][15] .…”

“…Previous studies employing deep learning architectures have predominantly used haemorrhage detection methods based on labelling or segmentation by experts 7,8,10,11,13,[15][16][17] . However, the classification of AIH is contingent on the opinion of experts, and the training of the system depends on the labelling of AIHsuspected areas by experts.…”