TumorSight Viz

TumorSight Viz is a software medical device designed to assist clinicians by providing advanced visualization and quantitative analysis tools for breast DCE-MRI studies in patients with early-stage or locally advanced breast cancer. Its methodology relies on processing DICOM images using features like automated segmentation and 3D rendering, with performance validated against radiologist-established ground truth to ensure clinically acceptable accuracy for supporting, but not solely determining, patient management decisions.

TumorSight Viz is a medical image management and processing system designed for visualizing and analyzing breast Magnetic Resonance Imaging (MRI) studies in patients with biopsy-proven early-stage or locally advanced breast cancer. It processes DICOM-formatted dynamic contrast-enhanced MRI (DCE-MRI) data, offering features such as image registration, subtractions, kinetic curve analysis, parametric image mapping, automated segmentation, 3D volume rendering, and various measurement tools. The system is a cloud-based, proprietary software accessed via a web application, intended to support clinicians in evaluating breast MRI findings, though it is explicitly stated that patient management decisions should not be made solely based on its results. The performance of TumorSight Viz was validated using an independent dataset of 161-163 breast cancer patient samples, with ground truth established by a consensus of U.S. Board Certified radiologists. This validation assessed the accuracy of automated measurements—such as tumor volume (mean absolute error [MAE] 6.48 cc), longest dimension (MAE 1.48 cm), and tumor-to-skin distance (MAE 0.63 cm)—and segmentation quality, achieving a mean volumetric Dice coefficient of 0.676 and a surface Dice coefficient of 0.873. The study concluded that the device's measurement error is consistent with inter-radiologist variability, demonstrating clinically acceptable performance for its intended use in visualizing and analyzing breast MRI studies.

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 deep learning algorithm for segmentation handle variability in image quality, MRI scanner types, and contrast administration protocols not extensively covered in the validation dataset?; What is the clinical impact of the reported mean absolute errors (e.g., 1.48 cm for longest dimension) on specific treatment planning decisions, such as surgical margin assessment or radiotherapy planning?; Given the mean volumetric Dice coefficient of 0.676, what are the common failure modes or types of tumors where the segmentation accuracy might be lower, and how are these flagged to the user?; Since patient management decisions should not be based solely on TumorSight Viz, what are the recommended workflows for integrating its outputs with other clinical information and radiologist expertise?; Are there plans to expand the functionality to include features present in the predicate device, such as interventional planning or serial comparison capabilities, and how might this impact its performance metrics?