{
  "abstract": "Background Acute PE is a life-threatening condition requiring prompt diagnosis and treatment. Whilst CTPA has improved PE detection, diagnostic accuracy remains limited by interobserver variability and suboptimal image acquisition, which contributes to under- and over-diagnosis. These errors may lead to inappropriate anticoagulation or missed treatment opportunities.FDA-approved AI software for acute PE detection is increasingly deployed in clinical workflows. However, validation often relies on comparison with initial radiology reports, which may themselves contain diagnostic inaccuracies. The AID-PE study (NCT06093217) addresses this limitation by using expert thoracic radiologists and senior clinicians to establish reference standards and clinical outcomes – believed to be one of the first such study to do so.Purpose To assess whether integrated AI software would improve diagnostic accuracy among general radiologists and reduce PE-related clinical risk in a single-centre DGH.Methods A subset of consecutive CTPAs from the AID-PE comparator cohort was retrospectively assessed using automated PE detection software. A senior respiratory clinician reviewed each initial radiology report determining whether scans were positive or negative for acute PE. The imaging reference standard was determined by independent expert thoracic radiologist review, blinded to the initial report and AI output.Results 664 CTPAs were re-evaluated (mean age 63 ± 19; 310 males, 354 females). Expert review detected acute PE in17.2% (114/664) of the original scans. Compared to expert review, the initial report demonstrated a sensitivity of 88.7% (95% CI 81.7–93.7%) and specificity of 98.7% (95% CI 97.8–99.6%). AI achieved slightly higher sensitivity at 90.4% (95% CI 84.9–95.8%) and comparable specificity at 98.5% (95% CI 97.5–99.5%).The initial report false negative rate was 12.3% (14/114) with AI correctly identifying 3 of these missed cases. The initial report false positive rate 1.3% (7/550) with AI correctly reclassifying 6 as negative. Two patients with under-reported PE experienced further VTE events, whilst one major bleeding episode occurred following treatment of an initial report false positive.Conclusions Preliminary results from the AID-PE study suggest AI closely aligns with expert interpretation than initial radiology reports. Findings highlight the potential for AI to improve diagnostic accuracy and reduce harm from misdiagnosis in acute PE.",
  "authors": [
    {
      "affiliations": [
        "Department of Respiratory Medicine, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK"
      ],
      "name": "JW Page"
    },
    {
      "affiliations": [
        "Department of Research and Development, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK"
      ],
      "name": "SGS Gunning"
    },
    {
      "affiliations": [
        "Department of Radiology, Royal United Hospitals Bath NHS Foundation Trust, Bristol, UK"
      ],
      "name": "PFP Charters"
    },
    {
      "affiliations": [
        "Department of Radiology, Royal United Hospitals Bath NHS Foundation Trust, Bristol, UK"
      ],
      "name": "S Lyen"
    },
    {
      "affiliations": [
        "Department of Radiology, Royal United Hospitals Bath NHS Foundation Trust, Bristol, UK"
      ],
      "name": "B Hudson"
    },
    {
      "affiliations": [
        "Department of Respiratory Medicine, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK"
      ],
      "name": "J Rossdale"
    },
    {
      "affiliations": [
        "Department of Research and Development, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK"
      ],
      "name": "A Seatter"
    },
    {
      "affiliations": [
        "Department of Respiratory Medicine, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK"
      ],
      "name": "R Mackenzie-ross"
    },
    {
      "affiliations": [
        "Department of Respiratory Medicine, Royal United Hospitals Bath NHS Foundation Trust, Bath, UK",
        "Department of Health, University of Bath, Bath, UK"
      ],
      "name": "J Suntharalingam"
    },
    {
      "affiliations": [
        "Department of Radiology, Royal United Hospitals Bath NHS Foundation Trust, Bristol, UK",
        "Department of Health, University of Bath, Bath, UK"
      ],
      "name": "JCL Rodrigues"
    }
  ],
  "title": "T1 Retrospective evaluation of artificial intelligence (AI) software for the detection of acute pulmonary embolism (PE) in CT pulmonary angiograms (CTPA): a preliminary ground-truth analysis from the AID-PE study",
  "uid": "90be1a91-8d43-59d6-958a-d53ae5018b1f"
}