Accuracy & clinical evidence

    How accurate is
    AI CT reporting?

    Published clinical studies and independent evaluations. Here's the evidence behind xAID's accuracy

    92.2%
    AI sensitivity across 9 pathology categories
    vs. 58.3% without AI assistance
    Pol Radiol, 2025
    95.6%
    AI specificity in emergency chest CT
    vs. 80.6% without AI assistance
    Pol Radiol, 2025
    94.9%
    Radiologists approved AI report elements for clinical integration
    Multi-center European study, 4 countries
    ResearchGate, 2025
    0.86
    Macro F1 score across clinically relevant pathologies
    Foundation model performance
    xAID technical evaluation

    Published clinical evidence

    Peer-Reviewed StudyPolish Radiology (Pol Radiol), 2025

    AI assistance in unenhanced chest CT: emergency setting evaluation

    A retrospective single-center study evaluated xAID Chest CT on 90 unenhanced chest CT scans in an emergency radiology setting. Pathologies assessed included lung nodules, opacifications, coronary calcification, aortic and pulmonary measurements, pleural and pericardial effusions, pneumothorax, and rib and vertebral fractures

    92.2%
    AI sensitivity
    vs. 58.3% unaided
    95.6%
    AI specificity
    vs. 80.6% unaided
    90
    CT scans
    evaluated
    9
    Pathology categories
    assessed

    AI outperformed radiologists in detection of coronary artery calcifications, pulmonary artery dilatation, and vertebral fractures

    Multi-Center StudyResearchGate, 2025 — 4 European centers

    xAID chest CT: retrospective clinical utility assessment

    A multi-center retrospective clinical utility assessment conducted across four European radiology centers (France, Greece, Slovakia, United Kingdom) evaluated xAID on 81 non-contrast chest CT cases with four board-certified radiologists

    94.9%
    Clinician approval
    for clinical integration
    89.7%
    Image layout approval
    across centers
    81.5%
    Diagnostic contribution
    of AI segmentation
    4
    European countries
    France, Greece, Slovakia, UK

    94.9% of radiologists agreed AI-generated structured report elements could be integrated into clinical practice with minor modifications

    Technical EvaluationxAID Foundation Model

    Foundation model performance

    xAID's core AI is a Swin transformer-based foundation model trained on large-scale clinical CT data. Input resolution up to 256³ voxels captures fine-grained anatomical structure across head, chest, and abdomen

    0.86
    Macro F1 score
    clinically relevant pathologies
    100+
    Findings analyzed
    per CT report
    256³
    Input resolution
    voxel-level analysis

    How xAID ensures report quality

    Multiple layers of verification — AI and human — before any report is delivered

    1

    Foundation model analysis

    A Swin transformer-based foundation model trained on clinical CT data analyzes the study. Input resolution up to 256³ voxels — capturing fine-grained anatomical detail

    2

    Secondary AI verification

    A second AI layer independently reviews the findings. Divergences between layers are flagged for radiologist attention

    3

    Radiologist review

    Every report is reviewed by a European radiologist before delivery. AI-assisted, not autonomous

    4

    95% accuracy — verified

    95% accuracy verified by peer-reviewed studies. Every report is reviewed by a radiologist before delivery

    95% Accuracy — Verified

    xAID achieves 95% accuracy verified by peer-reviewed studies. Published clinical evidence from independent research institutions confirms the accuracy of AI-assisted CT reporting with radiologist review

    Two independent peer-reviewed studies validate xAID's accuracy claims. Traditional teleradiology and narrow AI overlays offer no equivalent published clinical evidence

    Start a free 5-study pilot

    Accuracy questions

    In a peer-reviewed study published in Polish Radiology (2025), xAID achieved 92.2% pooled sensitivity and 95.6% specificity across 9 pathology categories in emergency chest CT — compared to 58.3% and 80.6% respectively for radiologists reviewing scans without AI assistance

    Published evidence suggests AI-assisted reading reduces missed findings rather than increasing them. In the Polish Radiology study, AI outperformed unaided radiologists in detecting coronary calcifications, pulmonary artery dilatation, and vertebral fractures. Every xAID report also receives radiologist review before delivery

    xAID achieves 95% accuracy verified by peer-reviewed studies. In a published study in Polish Radiology (2025), xAID achieved 92.2% sensitivity and 95.6% specificity. A separate multi-center European study found 94.9% of radiologists approved AI-generated report elements for clinical use

    xAID's core model is a Swin transformer-based foundation model with input resolution up to 256³ voxels. A second AI verification layer independently reviews findings before radiologist sign-off. The combined approach achieves a macro F1 score of 0.86 across clinically relevant pathologies

    Yes. The sensitivity/specificity figures come from a single-center retrospective study published in Polish Radiology (an independent peer-reviewed journal). The clinical utility data comes from a multi-center study across four European radiology centers. Neither was conducted by xAID internally