i2Contour

The i2Contour device is a web-based software platform that employs two distinct artificial intelligence models for the semi-automatic segmentation and volumetric quantification of WHO grade 4 glioblastoma from T1c and FLAIR MRI sequences. Its methodology involves AI-generated initial tumor contours which are then reviewed and manually refined by medical professionals, aiming to provide accurate and consistent tumor volume assessments to aid in managing patients 18 years of age or older with pathologically proven glioblastoma.

The MRIMath i2Contour is a web-based software platform designed to assist medical professionals in the analysis of WHO grade 4 glioblastoma (GBM) in adult patients aged 18 years or older. Its core purpose is the semi-automatic labeling, visualization, and volumetric quantification of GBM tumors derived from standard MRI images, specifically T1 post-contrast (T1c) and FLAIR sequences. The system utilizes two independent artificial intelligence (AI) algorithms—one dedicated to T1c sequences and another for FLAIR sequences—to generate initial tumor contours. These AI-generated segmentations are then presented to trained medical professionals through a user interface for review, manual adjustment if needed, and final approval. The i2Contour processes individual 2D slices and is fully automated, eliminating the need for pre-processing steps like skull stripping or image registration. It is explicitly stated that the software is an adjunctive tool, not intended for primary diagnosis or as the sole metric for diagnostic decisions, but rather to aid in the tumor contouring process and allow for comparison of volumetric measurements with past scans. Performance of the i2Contour software was evaluated by comparing its AI-generated contours against manual segmentations performed by three board-certified neuroradiologists on a dataset of 46 pre- and post-operative MRIs from patients with pathologically proven glioblastoma. For the FLAIR AI model, the Dice Similarity Coefficient (DSC) proportions exceeded a performance threshold of 0.88 in 85% of cases, achieving a mean overall DSC of 0.92, with median sensitivity and specificity of 93.4% and 98.6%, respectively. The T1c AI model showed that its DSC proportions surpassed the 0.88 threshold in 93% of instances, with a mean overall DSC of 0.95, and sensitivity and specificity of 92.7% and 97.2%, respectively. The evaluation also highlighted low inter-user variability when using the MRIMath smart manual contouring platform. Furthermore, the AI models produced mean Hausdorff distances under 5 mm and volumetric measurements, kappa scores, and Bland-Altman differences that closely matched those of the radiologists, indicating high accuracy for both T1c and FLAIR AI models in segmenting GBM.

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 i2Contour software account for variations in MRI image quality and acquisition parameters from different scanners and institutions in its AI segmentation?; What is the typical time saving for a neuroradiologist when using i2Contour for GBM contouring compared to fully manual segmentation, and how does this impact workflow?; Could the performance metrics (DSC, sensitivity, specificity) be stratified by specific GBM characteristics, such as tumor location, size, or post-treatment changes?; How does the 'smart manual contouring platform' component enhance the user's ability to refine AI-generated contours, and what specific tools does it offer?; Beyond volumetric measurements, are there plans to extract or correlate other quantitative imaging biomarkers using the i2Contour platform?