Pilot Report for Intracranial Hemorrhage Detection with Deep Learning Implanted Head Computed Tomography Images at Emergency Department

Abstract

Hemorrhagic stroke is a serious clinical condition that requires timely diagnosis. An artificial intelligence algorithm system called DeepCT can identify hemorrhagic lesions rapidly from non-contrast head computed tomography (NCCT) images and has received regulatory clearance. A non-controlled retrospective pilot clinical trial was conducted. Patients who received NCCT at the emergency department (ED) of Kaohsiung Veteran General Hospital were collected. From 2020 January-1^st to April-30^th, the physicians read NCCT images without DeepCT. From 2020May-1^st to August-31^st, the physicians were assisted by DeepCT. The length of ED stays (LOS) for the patients was collected. 2,999 patients were included (188 and 2811 with and without ICH). For patients with a final diagnosis of ICH, implementing DeepCT significantly shortened their LOS (560.67 ± 604.93 min with DeepCT vs. 780.83 ± 710.27 min without DeepCT; p = 0.0232). For patients with a non-ICH diagnosis, the LOS did not significantly differ (705.90 ± 760.86 min with DeepCT vs. 679.45 ± 681.97 min without DeepCT; p = 0.3362). For patients with ICH, those assisted with DeepCT had a significantly shorter LOS than those without DeepCT. For patients with a non-ICH diagnosis, implementing DeepCT did not affect the LOS, because emergency physicians need same efforts to identify the underlying problem(s) with or without DeepCT. In summary, implementing DeepCT system in the ED will save costs, decrease LOS, and accelerate patient flow; most importantly, it will improve the quality of care and increase the confidence and shorten the response time of the physicians and radiologists.

Keywords: Artificial intelligence (AI); Convolutional neural network; Cost efficacy; Deep learning; Head non-contrast computed tomography; Intra-cranial hemorrhage (ICH).