{
  "abstract": "Background Timely, accurate diagnosis of intracranial aneurysms (IA) is critical due to their risk of rupture and resultant life-threatening subarachnoid hemorrhage (SAH). As radiological workflow burden increases, AI-based methods for triage can enhance workflow efficiency and performance. This study investigates the impact of an AI algorithm on IA detection in a real-world clinical setting.Methods We included 2,466 consecutive adult CTA studies performed over five months (November 2023 to April 2024) in a large, integrated, multisite healthcare system. In this population, the AI algorithm was not yet implemented during the initial radiologist interpretation. Unaided radiologist diagnoses were extracted from the associated radiology report using natural language processing. A commercially available, FDA-cleared AI algorithm (Aidoc, Tel Aviv, Israel) then retrospectively evaluated all CTA studies and identified those with suspected IAs. Cases identified as negative for IA in the radiology report but positive by the AI algorithm underwent adjudication for ground truth by two interventional neuroradiologists. Enhanced detection was defined as AI-identified IAs not detected during initial interpretation.Results Initial radiologist interpretation identified 54 IA positive cases (2.2%). The AI algorithm flagged 29 additional cases, of which 22 were confirmed by adjudication to be true positives, equating to a 40.7% (22/54) increase in detected IAs. Of the 22 cases newly identified as positive by AI, 20 were located in the anterior circulation (12 ophthalmic/hypophyseal, 1 ACOM, 2 PCOM, 4 MCA, and 1 cavernous ICA), and 2 cases were located in the posterior circulation (1 vertebrobasilar junction, 1 basilar apex). The 7 AI false positives were evenly distributed in the anterior and posterior circulations. Of the 22 AI true positive cases not detected by the unaided radiologist, 5 (22.7%) were interpreted by fellowship-trained neuroradiologists and 17 (77.3%) were interpreted by other radiologists. Mean IA size in the 22 AI true positive cases was 4.2 ± 1.2mm.Conclusion The integration of an AI algorithm for IA triage into the radiological workflow can improve IA detection. The ophthalmic segment was the most common site of enhanced aneurysm detection by AI. The distribution of initial readers suggests that aneurysm detection AI tools may be more beneficial to radiologists without neuroradiology fellowship training.Disclosures J. Milburn: 2; C; AI Doc. P. Gulotta: None. V. Loving: None. M.C. Morvant: None. A. Toshav: None.",
  "authors": [
    {
      "affiliations": [
        "Radiology, Ochsner Medical Center, Jefferson, LA"
      ],
      "name": "J Milburn"
    },
    {
      "affiliations": [
        "Radiology, Ochsner Medical Center, Jefferson, LA"
      ],
      "name": "P Gulotta"
    },
    {
      "affiliations": [
        "Radiology, Ochsner Medical Center, Jefferson, LA"
      ],
      "name": "V Loving"
    },
    {
      "affiliations": [
        "Radiology, Ochsner Medical Center, Jefferson, LA"
      ],
      "name": "MC Morvant"
    },
    {
      "affiliations": [
        "Radiology, Ochsner Medical Center, Jefferson, LA"
      ],
      "name": "A Toshav"
    }
  ],
  "title": "E-342 Improving intracranial aneurysm detection with an AI-enhanced workflow",
  "uid": "b6daedf9-a329-5395-b1d8-29fea2dd6cff"
}