Medical-AI trial: non-invasive AI detection of IDH-wildtype glioblastoma from MRI (GliomaAI-GBM, brain tumor, 1,372 public scans) (NCT07673692)
An observational study evaluating whether an AI system, GliomaAI-GBM, can identify the molecular subtype IDH-wildtype glioblastoma from routine MRI. It uses 1,372 fully anonymized scans from 13 institutions in the Cancer Imaging Archive (TCIA) and measures accuracy, sensitivity, specificity, and area under the ROC curve. Completed.
Trial overview (primary data)
- StatusCompleted
- ConditionsGlioma, Glioma (Diagnosis), Glioblastom WHO Grade 4, IDH Wildtype Glioblastoma
- InterventionsDEVICE: GliomaAI-GBM
- SponsorDeep Learning Institute of Radiological Sciences
- Target enrollment1,372 participants
- Period2017-03-14 〜 2021-07-16
Key points
- An observational study of whether AI (GliomaAI-GBM) can non-invasively identify IDH-wildtype glioblastoma from routine MRI.
- Data are 1,372 anonymized MRI scans from 13 institutions in the Cancer Imaging Archive (TCIA); retrospective, with no new scans, treatment, or visits.
- Primary endpoint is diagnostic performance (accuracy, sensitivity, specificity, predictive values, area under the ROC), with generalization across hospitals and patient groups.
- IDH status is a key marker for glioma classification, prognosis, and treatment, but normally needs biopsy and molecular testing.
- With sibling studies (astrocytoma, oligodendroglioma) it spans the major subtypes; it does not establish clinical use or biopsy replacement.
Glioblastoma is among the most aggressive brain tumors. In current classification, whether a tumor carries a mutation in the IDH gene (IDH-mutant) or not (IDH-wildtype) is a key marker that shapes the diagnosis, prognosis, and treatment plan. This subtype is normally determined by biopsy and subsequent molecular testing, but brain biopsy carries burden and risk.
If AI could estimate the subtype from MRI images alone, it might provide information earlier and less invasively.
This trial applies AI (GliomaAI-GBM) to that non-invasive task as an observational study.
Per the registry summary, it uses anonymized MRI and clinical information from 1,372 patients across 13 institutions in the Cancer Imaging Archive (TCIA), a public research database, to measure how accurately GliomaAI-GBM identifies IDH-wildtype glioblastoma (accuracy, sensitivity, specificity, positive/negative predictive value, area under the ROC curve) and how well it generalizes across different hospitals and patient groups.
It is retrospective, asks nothing of participants, and uses only fully anonymized images. The same research program also registers sibling studies for astrocytoma (grades 2, 3, and 4) and oligodendroglioma, covering the major molecular subtypes of glioma with MRI and AI.
IDH status is central to glioma classification and prognosis, yet confirmation typically requires tissue testing. Estimating it non-invasively from MRI with AI is an example of radiogenomics, reading genetic features from radiology images, and evaluating generalization across multiple institutions is a design geared toward real deployment.
That said, this is a retrospective diagnostic-performance study on existing images; it does not establish clinical use or replacement of biopsy.
Why it matters
Radiogenomics, estimating genetic features from radiology images, is an area expected to reduce invasiveness and yield earlier information. A design that evaluates generalization across multiple institutions is a useful reference for deploying and evaluating imaging AI (this trial is a retrospective performance study, not an establishment of clinical use).
FAQ
Why does IDH status (wildtype vs. mutant) matter?
Can MRI alone replace a biopsy?
Sources (primary)
Source: ClinicalTrials.gov (U.S. NIH/NLM, public domain). This site does not provide medical advice. Verify the latest and exact details with the official source. This site is not endorsed or certified by the NIH/NLM.
- ClinicalTrials.gov (study record, original)
- NCT ID: NCT07673692