i2Contour

K233822

MRIMath LLC · cleared 2024-08-08 · product code QIH · Radiology

Premarket evidence — what FDA accepted

Device typesamd

source quote (p.5)

“The MRIMath i2Contour is a web-based software platform designed for the contouring and segmentation of the T1c and FLAIR sequences of the MRIs of patients already diagnosed with GBM.”

AlgorithmAI-powered segmentation of magnetic resonance images (MRI) using two independent AIs (one for T1c and one for FLAIR series), fully automated, processes individual 2D slices.

source quote (p.8)

“AI-powered segmentation of the magnetic resonance images (MRI) of patients diagnosed with glioblastoma multiforme is the technological principle for both the subject and predicate devices. The subject device includes two independent AIs, one for T1c and the other for the FLAIR series; the predicate device consists of a single AI. The subject device processes individual 2D slices. The subject device is fully automated as it does need registration nor skull stripping.”

Adaptive (vs locked)FDA source did not state this

PCCPFDA source did not state this

Cybersecurity addressedFDA source did not state this

Validation studies (2)

Retrospective clinical

n=33 patients

endpoints: mean DSC

Retrospective clinical

n=46 images · 19 site(s)

endpoints: accuracy of AI contours; mean DSC; sensitivity; specificity; mean Hausdorff distances; volume measurements; kappa scores; Bland-Altman differences

Reported performance (9 observations)

sensitivity0.927

source quote (p.9)

“Sensitivity and specificity for T1c AI were 92.7% and 97.2%, respectively.”

specificity0.972

source quote (p.9)

“Sensitivity and specificity for T1c AI were 92.7% and 97.2%, respectively.”

diceas written: “Mean overall DICE score for T1c AI”0.95CI (93%, 96%)

source quote (p.9)

“The mean overall DICE scores for the post-contrast T1 (T1c) AI were 0.95 with a 95% confidence interval (C.I) of (93%, 96%), closely matching the radiologists' scores.”

diceas written: “Mean DSC for true positive T1c images”0.83

source quote (p.9)

“For true positive T1c images, AI segmentation scored a mean DSC of 83%, versus radiologists' ranging from 76% to 86%.”

diceas written: “FLAIR AI mean DSC”0.92CI (90%, 94%)

source quote (p.9)

“The FLAIR AI mean DSC was 92% with a 95% CI interval of (90%, 94%), also matching the radiologists scores.”

diceas written: “Mean DICE score for true positive FLAIR slices”0.8

source quote (p.9)

“The AI also achieved a mean DICE score of 80% for true positive FLAIR slices, against the radiologists' 75%-83%”

sensitivityas written: “Median sensitivity for FLAIR AI”0.934

source quote (p.9)

“and exhibited a median sensitivity and specificity of 93.4% and 98.6%, respectively.”

specificityas written: “Median specificity for FLAIR AI”0.986

source quote (p.9)

“and exhibited a median sensitivity and specificity of 93.4% and 98.6%, respectively.”

agreement_kappaas written: “Kappa scores”stated without value

source quote (p.9)

“The T1C and FLAIR AI models also produced mean Hausdorff distances (< 5 mm), volume measurements, kappa scores, and Bland-Altman differences that align closely with measurements by radiologists.”

Each value carries its own analysis unit and task — never compare or pool across devices. Source: 510(k) summary PDF.

Predicate network

Postmarket — what happened after clearance

recalls in product code, 24mo

MAUDE reports in code, 12mo

—

vs code's own 3-yr baseline

drift signals on this device

Recall and MAUDE counts are product-code-level (reports aren't reliably attributable to one device). Signals are descriptive observables with sources — never a judgment that the device is unsafe or drifting. Snapshot 2026-07-08.

Reimbursement — how devices like this got paid

Not yet tracked — no payment pathway indexed for this clearance (the reimbursement corpus is a growing seed set).