EUVIMEDEUVIMED entdecken
Crossref

Therapeutic Hybrid Intelligence with Neural and Knowledge-based Expert Reasoning for SRS (THINKERS): a mixture-of-experts AI model for vestibular schwannoma

Jheremy S Reyes, Constantinos G Hadjipanayis, Ajay Niranjan

Journal of Neuro-Oncology · 2026-07 · 178 (3)

DOI: 10.1007/s11060-026-05679-0 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Neuro-Oncology, online erschienen am 18.06.2026. Verfasst von Jheremy S Reyes und 2 weiteren Autor:innen. Beteiligte Einrichtung: Center for Image-Guided Neurosurgery, Department of Neurological Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. Hauptthemen laut MEDLINE: Neuroma, Acoustic, Radiosurgery, Artificial Intelligence, Neural Networks, Computer. Die Arbeit wurde bisher 14-mal zitiert (Crossref) und nennt 13 Literaturangaben.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

Outcome prediction after Gamma Knife radiosurgery (GKRS) for vestibular schwannoma remains largely guided by tumor size, Koos grade, baseline symptoms, cochlear dose constraints, and institutional experience rather than individualized estimates of tumor progression and functional outcomes. We developed THINKERS-VS, a mixture-of-experts (MoE) artificial intelligence framework for progression and symptom-transition prediction after GKRS.

Methods

We performed a retrospective single-center study of vestibular schwannomas treated with GKRS. Variables available before the treatment were used to train a MoE neural network with discrete-time survival modeling. The model incorporated demographic, clinical, tumor, and radiosurgical variables. The primary endpoint was tumor progression across intervals of 3, 6, 12, 18, 24, 32, 48, and 60 months. Secondary endpoints included new or worsened hearing loss, imbalance, vertigo, dizziness, and tinnitus. Internal validation used grouped 5-fold cross-validation and a grouped holdout test split. Performance was assessed using AUC and Brier score.

Results

The cohort included 686 patients. Median age was 59.5 years, median tumor volume was 0.746 cm³, median prescription dose was 12.0 Gy, and median follow-up was 52.3 months. THINKERS-VS achieved strong progression discrimination across evaluated intervals, with AUCs ranging from 0.807 to 0.859. At 60 months, AUC was 0.807 and Brier score was 0.012. For symptom-transition prediction, holdout AUCs were 0.875 for hearing loss, 0.844 for imbalance, 0.813 for vertigo, 0.884 for dizziness, and 0.839 for tinnitus.

Conclusions

THINKERS-VS provides an internally validated framework for individualized tumor progression and symptom-transition prediction after GKRS for vestibular schwannoma and recommends tumor margin dose associated with best outcome. Clinical trial number Not applicable.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Neuroma, AcousticRadiosurgeryArtificial IntelligenceNeural Networks, ComputerHumansFemaleRetrospective StudiesMaleMiddle AgedAgedAdultDisease ProgressionTreatment OutcomePrognosisFollow-Up Studies

Stichwörter der Autor:innen

Artificial intelligenceGamma Knife radiosurgeryMixture-of-expertsOutcomes predictionStereotactic radiosurgeryVestibular schwannoma

Autor:innen und Institutionen

  • Center for Image-Guided Neurosurgery, Department of Neurological Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA, USA.
  • Computational Neurosurgery Research Group (CIGNS-CRG), Center for Image-Guided Neurosurgery, Department of Neurological Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA, USA.
  • Center for Image-Guided Neurosurgery, Department of Neurological Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. niraax@upmc.edu.
  • Computational Neurosurgery Research Group (CIGNS-CRG), Center for Image-Guided Neurosurgery, Department of Neurological Surgery, University of Pittsburgh Medical Center, Pittsburgh, PA, USA. niraax@upmc.edu.

Zitieren

APA

Reyes, J. S., Hadjipanayis, C. G., & Niranjan, A. (2026). Therapeutic Hybrid Intelligence with Neural and Knowledge-based Expert Reasoning for SRS (THINKERS): a mixture-of-experts AI model for vestibular schwannoma. Journal of Neuro-Oncology, 178(3). https://doi.org/10.1007/s11060-026-05679-0

Vancouver

Reyes JS, Hadjipanayis CG, Niranjan A. Therapeutic Hybrid Intelligence with Neural and Knowledge-based Expert Reasoning for SRS (THINKERS): a mixture-of-experts AI model for vestibular schwannoma. Journal of Neuro-Oncology. 2026;178(3). doi:10.1007/s11060-026-05679-0

BibTeX
@article{Reyes2026,
  author = {Reyes, Jheremy S and Hadjipanayis, Constantinos G and Niranjan, Ajay},
  title = {Therapeutic Hybrid Intelligence with Neural and Knowledge-based Expert Reasoning for SRS (THINKERS): a mixture-of-experts AI model for vestibular schwannoma},
  journal = {Journal of Neuro-Oncology},
  year = {2026},
  volume = {178},
  number = {3},
  doi = {10.1007/s11060-026-05679-0}
}
Kostenlos starten →