Truace
Subscribe

Interim study of 70 patients finds imaging plus AI can tell meningioma from healthy brain

Study title: Machine learning-assisted intraoperative hyperspectral imaging in cranial meningioma surgery: interim results of an ongoing study

An interim analysis of an ongoing prospective observational study tested whether hyperspectral imaging, a label-free optical technique, combined with machine learning could tell cranial meningioma from nearby tissue during surgery. Among 70 patients, the best classifier separated meningioma from cortex with an accuracy of 0.928.

Read the study abstract

Objective Hyperspectral imaging (HSI) is a label-free optical technique that might extend conventional intraoperative visualization by providing quantitative spectral information from the operative field. The aim of this study was to evaluate whether intraoperative in vivo HSI combined with machine learning (ML)-based analysis can distinguish cranial meningioma from adjacent intracranial tissues, particularly cortex and dura, during meningioma surgery. Methods This interim analysis of an ongoing prospective obse…

Truvace record2 min read1 sourceEvidence: Moderate · 56/100 — Observational study · n = 70 · 1 independent sourcePeer-reviewedTRV-2026-2432 · cite or check it
Machine learning-assisted intraoperative hyperspectral imaging in cranial meningioma surgery: interim results of an ongoing study"Medical Laboratory" by ben.dracup, CC BY 2.0.

In brief

Meningioma showed distinct spectral features, including a weaker oxyhemoglobin-associated double peak and lower tissue oxygenation than cortex and dura. When the task widened to three tissue types (meningioma, cortex and dura), accuracy fell to 0.773 and macro F1 to 0.782.

The authors call the findings promising for label-free optical guidance but say external validation, patient-level analyses, WHO grade prediction and spatially matched histopathological confirmation are needed before clinical use.

Main points

  1. The study included 70 patients having elective cranial meningioma surgery at a single tertiary neurosurgical center.
  2. Researchers analyzed 291,104 meningioma pixels from 248 regions, 141,515 cortex pixels from 135 regions and 333,244 dura pixels from 104 regions.
  3. For meningioma versus cortex, the multilayer perceptron reached precision 0.960, recall 0.932, F1 0.946 and a Matthews correlation coefficient of 0.840.
  4. In the three-class task, macro F1 was 0.782 and macro Matthews correlation coefficient 0.662.
  5. Meningioma showed lower tissue oxygenation than cortex and dura.
  6. Four classifiers were tested using leave-one-patient-out cross-validation.

The gain

In an interim study of 70 meningioma surgery patients, a machine learning classifier applied to hyperspectral imaging distinguished tumour from cortex with 0.928 accuracy, and from cortex and dura with 0.773 accuracy.

The full story

Imaging was done with a CE-marked tissue system at predefined surgical time points. Experts annotated regions of interest for meningioma, cortex and dura on HSI-derived RGB images, and pixel-wise reflectance spectra were extracted from them.

The team compared raw spectra, standard normal variate-normalized spectra and first- and second-derivative spectra across four classifiers: linear discriminant analysis, support vector machine, logistic regression and multilayer perceptron. Leave-one-patient-out cross-validation was used to evaluate them.

The multilayer perceptron on normalized spectra performed best in both tasks. Performance dropped in the multiclass task, with accuracy of 0.773 versus 0.928 for the binary task, though the authors describe it as feasible.

Sources

  1. Peer-reviewedNeurosurgical Focus2026-11-01

The debate