VisiRad XR

VisiRad XR is a machine learning-driven software designed to enhance the detection of pulmonary nodules and masses (6-60mm) on adult chest radiographs by serving as a second-read aid for clinicians, processing DICOM images through a cloud-based system. Clinical evaluations confirmed its ability to improve radiologists' diagnostic accuracy (AUC) and sensitivity, despite a noted decrease in specificity, in identifying these lung abnormalities.

VisiRad XR, from Imidex Inc., is a Software as a Medical Device (SaMD) utilizing machine learning for computer-aided detection (CADe) of lung nodules and masses, ranging from 6mm to 60mm, on chest radiographs. It processes DICOM images (both AP and PA views from various hardware systems) via a secure cloud-based platform, outputting a secondary DICOM with annotated regions of interest (ROIs). This device is intended as a second-read tool for physicians, aiding in the identification of suspicious findings in adult patients (>21 years) whose radiographs were taken in outpatient or emergency department settings; it is not designed to replace the radiologist's initial interpretation, nor is it for triage or definitive diagnosis. The device's efficacy was evaluated through comprehensive non-clinical and clinical studies. Standalone performance testing on a broad dataset yielded an overall sensitivity of 0.83 with an average of 1.5 false positives per image and an Area Under the ROC Curve (AUC) of 0.73. A multi-reader, multi-case (MRMC) retrospective clinical study with 24 radiologists and 600 images demonstrated that VisiRad XR significantly improved reader accuracy, achieving a mean increase in AUC of 0.027 when used as an aid compared to unaided reads. This was accompanied by an average increase in sensitivity of 0.076, though average specificity decreased by 0.086. The software carries a Moderate level of concern, indicating that a malfunction could potentially lead to minor injury from an erroneous diagnosis or delayed medical care.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the performance of VisiRad XR, particularly its sensitivity and false positive rate, vary with nodule/mass size specifically at the lower (6mm) and upper (60mm) ends of its detection range and across different radiographic views (AP vs. PA)?; What specific machine learning architectural choices or training data characteristics contribute to the observed 0.086 decrease in specificity when using VisiRad XR, and what are the clinical implications of this trade-off with increased sensitivity?; Given VisiRad XR's cloud-based architecture, what are the detailed data security protocols, including encryption methods and compliance with privacy regulations like HIPAA, during DICOM image transfer, processing, and storage?; The MRMC study simulated clinical practice by sequential reads (unaided then aided). How might real-world integration, where VisiRad XR results are available concurrently with or prior to the initial read, affect radiologist workflow, reading times, and diagnostic confidence over extended periods of use?; What was the spectrum of clinical experience (e.g., years in practice, subspecialty) among the 24 radiologists in the MRMC study, and did the magnitude of improvement in AUC and sensitivity with VisiRad XR show any correlation with these reader characteristics?