{
  "abstract": "Background and Purpose Artificial intelligence (AI)-based intracranial aneurysm detection software has recently been introduced and is increasingly used as a screening tool in clinical practice. While prior studies have primarily focused on algorithm performance and diagnostic accuracy, clinical efficacy also depends on the system’s ability to successfully process imaging data and generate interpretable output. Recent reports suggest that a substantial proportion of examinations encounter technical failures. Despite their potential clinical impact, the incidence and underlying mechanisms of these failures remain poorly understood. Addressing these limitations may further enhance the clinical effectiveness of AI-based tools. Using a widely adopted platform (Viz ™ Aneurysm), we evaluated the types of technical failures and analyzed their underlying mechanisms.Materials and Methods In this single-center retrospective study, all consecutive head CTA examinations performed over a 6-month period were included regardless of clinical indication. The results of Viz ™ Aneurysm for each examination (aneurysm-positive or negative) were compared with radiologist interpretations as the reference standard. Diagnostic performance was evaluated at the patient level, with aneurysm-level subgroup analyses. Examinations that failed to produce interpretable AI output due to technical failure were analyzed separately to determine their frequency and contributing mechanisms and were excluded from primary performance calculations.Results Among 1,331 patients, 134 (10.1%) were diagnosed with intracranial aneurysms based on radiologist interpretations, yielding a total of 177 aneurysms. When a 4-mm size threshold was applied, sensitivity and specificity were 87.4% and 98.6%, respectively. Notably, 37 of 177 aneurysm-positive examinations (21.6%) failed to generate interpretable AI output due to technical issues, including 9 data transfer failures (DTF) and 28 series processing failure (SPF) cases. Analysis of these technical failures demonstrated multiple contributing mechanisms across the imaging and system workflow. Among SPF cases, 14 (50.0% of SPF; 37.8% of all failures) were associated with metal-related artifacts, which were often caused by devices unrelated to the region of interest, such as ventriculoperitoneal shunts, external ventricular drains, or contralateral surgical hardware. The remaining 14 SPF cases (50.0% of SPF; 37.8% of all failures) were attributed to multifactorial causes, likely reflecting image quality variability, preprocessing limitations, and other system-level constraints.Conclusions AI-based aneurysm detection demonstrated performance comparable to prior reports when clinically relevant thresholds were applied. However, a substantial proportion of examinations failed to generate interpretable output due to technical failures, including data transfer failure (DTF) and series processing failure (SPF). These findings highlight the need for system-level optimization beyond algorithmic accuracy, which may further enhance the clinical effectiveness of AI-based aneurysm detection tools.Disclosures I. Yuki: None. K. Aoki: None. R. Back: None. J. Carbone: None. T. Nguyen: None. C. Jin: None. B. Malvar: None. K. Golshani: None. F. Hsu: None. S. Suzuki: None. S. Suzuki: None. J. Soun: None.",
  "authors": [
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "I Yuki"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "K Aoki"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "R Back"
    },
    {
      "affiliations": [
        "Boston Children’s hospital, Boston, MA"
      ],
      "name": "J Xu"
    },
    {
      "affiliations": [
        "Radiology, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "J Carbone"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "T Nguyen"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "C Jin"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "B Malvar"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "K Golshani"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "F Hsu"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "S Suzuki"
    },
    {
      "affiliations": [
        "Neurosurgerey, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "S Suzuki"
    },
    {
      "affiliations": [
        "Radiology, UC Irvine Medical Center, Orange, CA"
      ],
      "name": "J Soun"
    }
  ],
  "title": "E-300 Beyond diagnostic accuracy: evaluating technical failure mechanisms in AI-based intracranial aneurysms detection",
  "uid": "d8ade750-3b0c-515e-8715-e8564855160b"
}