E-078 Advance study: automated detection and volumetric assessment of intracerebral hemorrhage. (22nd July 2018)

• Record Type: Journal Article
• Authors: Barreira, C; Bouslama, M; Ratcliff, J; Pradilla, G; Rahman, H; Al-Bayati, A; Haussen, D; Grossberg, J; Frankel, M; Nogueira, R
• Is Part Of: Journal of neurointerventional surgery. Volume 10(2018)Supplement 2
• Page Start: A88
• Page End: A88
• Publication Date: 2018-07-22
• Subjects: Nervous system -- Surgery -- Periodicals; Cerebrovascular disease -- Surgery -- Periodicals; 617.48
• DOI: 10.1136/neurintsurg-2018-SNIS.154
• Languages: English
• ISSNs: 1759-8478

Abstract

Abstract : Introduction: Intracerebral hemorrhages (ICHs) are a source of significant mortality and morbidity worldwide, and even those who survive the insult have a higher future mortality than the general population. 1 Given this, prompt recognition, hematoma volume and location, differentiation from intraventricular hemorrhage (IVH) and the presence of hydrocephalus are known features related to unfavorable outcomes. An artificial intelligence (AI) fully automated detection system is thought to ameliorate ICH detection and measurement, which could potentially improve condition recognition and patient prognosis. Methods: A retrospective analysis of non-contrast CTs (NCCTs) from a single center, randomly picked from prospective cohort of acute stroke patients, was done to compare the semi-automated expert-based (OsiriX MD v.9.0.1) ratings versus AI algorithm ratings. NCCTs from 2014 to 2017, for both ICH (presence and volume) and other neuroimaging findings (such as IVH) and non-ICH subjects were analyzed. Same scans were rated by Viz-ICH® v2.0 – an Artificial Intelligence Automated Convolutional Neural Network. Results: Preliminary analysis of subjects: 689 NCCTs were evaluated – of those, 537 had ICH and 152 were non-ICH (control group). ICH group findings: mean age 59.3±13.5, bNIHSS 13[4–22], ICH volume of 26.6±36 cc, males 56.4% hypertension 86.2% and presence of IVH 51.7% of them. Intraclass Correlation Coefficient (uncontrolled for IVH): α=96.1% (IC95%=0.900–0.933; p≤0.001) and sensitivity 94.5%. Maximal running time of the algorithm was under 15s. Conclusion: The presence and volume of ICHs can be accurately predicted by AI Viz-ICH Algorithm, with good differentiation from IVH and other neuroimaging findings. The software has the potential for appropriate and rapid diagnosis of these patients. Disclosures: C. Barreira: None. M. Bouslama: None. J. Ratcliff: None. G. Pradilla: None. H. Rahman: None. A. Al-Bayati: None. D. Haussen: None. J. Grossberg: None. M. Frankel: None. R. Nogueira: None.