Purpose:
To evaluate the diagnostic performance of a commercially available AI tool for detecting ICH on non-contrast head CT, using expert radiologist consensus as the reference standard.
Materials and Methods:
This retrospective study included 973 anonymized non-contrast head CT scans from routine clinical practice, representing a mix of hemorrhagic and non-hemorrhagic cases. The AI tool was tested for its ability to detect brain hemorrhages, with ground truth labels determined by consensus review from two board-certified neuroradiologists. Diagnostic performance was assessed using the area under the receiver operating characteristic curve (AUC).
Results:
The AI tool demonstrated strong diagnostic performance across various hemorrhage subtypes, with high AUC values: 0.885 for intracranial hemorrhage, 0.905 for pericerebral hematoma, 0.898 for subdural hematoma, 0.827 for subarachnoid hemorrhage, 0.918 for intraparenchymal hemorrhage, and 0.948 for intraventricular hemorrhage. The tool showed high specificity (>0.94), making it effective for ruling out hemorrhages. Sensitivity varied from 0.43 (subdural hematoma) to 0.79 (intracranial hemorrhage), with potential for improvement through post-deployment threshold adjustments. Negative predictive value (NPV) was high (0.71–0.98), supporting triage use, while positive predictive value (PPV) ranged from 0.46 to 0.90, highlighting the need for radiologist confirmation.
Conclusion:
The AI tool demonstrated strong performance, with high specificity and NPV across multiple hemorrhage subtypes. Its ability to detect both common and critical hemorrhages supports its role in emergency and high-volume settings. Adjusting thresholds could improve sensitivity, optimizing its clinical utility.
Clinical Relevance:
AI-assisted CT interpretation can expedite diagnosis, reduce missed findings, and assist radiologists in delivering timely care, particularly in high-demand environments. It can also reduce clinician workload and improve patient outcomes by aiding early intervention in cases of brain haemorrhage.
