Truace
Subscribe
60/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Model flags brain vessel malformations in 142 HHT patients; wider findings less reliable

Study title: A machine learning approach to MRI screening for brain arteriovenous malformations in hereditary hemorrhagic telangiectasia

Evidence type
Observational study
Sample
n = 142
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2476cite or check it

Study title: A machine learning approach to MRI screening for brain arteriovenous malformations in hereditary hemorrhagic telangiectasia

Researchers built two machine learning models, logistic regression and random forest, to predict brain arteriovenous malformations (bAVMs) in 142 patients with confirmed hereditary hemorrhagic telangiectasia. Both reached good discrimination, with an AUC of 0.84 for logistic regression and 0.85 for random forest.

Read the study abstract

Background Hereditary hemorrhagic telangiectasia (HHT) is an inherited vascular disorder characterized by telangiectasias and arteriovenous malformations (AVMs) in multiple organs. Although brain MRI screening is recommended in all patients with HHT, in many centers, this may be difficult to implement due to organizational constraints. The aim of this study was to develop a machine learning model to predict the presence of bAVMs and other cerebrovascular malformations in patients with HHT, with the goal of suppo…

The gain

In a study of 142 patients with hereditary hemorrhagic telangiectasia, machine learning models predicted brain arteriovenous malformations with an AUC of 0.84 (logistic regression) and 0.85 (random forest).

In brief

Brain MRI screening is recommended for all HHT patients, but many centers find it hard to carry out because of organizational constraints. The authors aim to support risk stratification in centers that already hold genotype information and to identify patients who may suit closer surveillance.

The broader model did not hold up: those predicting any cerebrovascular malformation showed limited discrimination and poor calibration. The authors say the bAVM tool needs external validation before it could guide how screening is allocated.

Main points

  1. The study included 142 patients with confirmed HHT who met at least 3 Curaçao criteria and had complete genetic work-up and radiological screening between 2018 and 2026.
  2. Logistic regression reached an AUC of 0.84 (95% CI 0.73-0.94) for bAVM prediction; random forest reached 0.85 (95% CI 0.74-0.96).
  3. Random forest had higher sensitivity (0.73 vs 0.69) but lower specificity, PPV and F1 score, with the same NPV.
  4. Logistic regression was better calibrated, with an intercept of -0.30 and slope of 0.87, versus 1.39 and 2.39 for random forest.
  5. Models for any cerebrovascular malformation showed limited discrimination and poor calibration.
  6. Top logistic regression features were deletion mutations, ENG, family history, ACVRL1 and truncated variants.

The full story

The team defined two end points: the probability of detecting bAVMs and the probability of detecting any cerebrovascular finding. Analysis used Python and R, and patients were drawn from those with complete genetic and radiological data between 2018 and 2026.

For bAVM prediction the two models differed in trade-offs. Random forest had a slightly higher AUC and sensitivity (0.73 vs 0.69), while logistic regression had better specificity, PPV, F1 score and calibration.

Feature analysis pointed to genotype and family history. Logistic regression ranked deletion mutations, ENG, family history, ACVRL1 and truncated variants highest, while random forest SHAP analysis ranked ACVRL1, ENG, age, pulmonary AVMs and family history.

The authors conclude the phenotype-specific bAVM model achieved effective risk stratification, and that after external validation it may allow more efficient allocation of screening strategies.

Talk it overFour questions for a class, a team or the dinner table

  1. Who gets this benefit, and who was left out of the study or story?

    The gain In a study of 142 patients with hereditary hemorrhagic telangiectasia, machine learning models predicted brain arteriovenous malformations with an AUC of 0.84 (logistic regression) and 0.85 (random forest).

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on an observational study with a sample of 142, from Neuroradiology.

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The bAVM tool has not been externally validated, and the models for any cerebrovascular malformation showed limited discrimination and poor calibration.

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

The questions are the same on every record. The details under them are taken from this record as written.

Sources

1 on file
  1. Neuroradiologydoi:10.1007/s00234-026-04253-9
    Peer-reviewed2026-11-07

The debate

More from Health