A Vision-Language Foundation Model for Brain Disease Diagnosis From Multimodal Data

Xiangya Hospital of Central South University100,000 enrolled

Overview

The goal of this observational study is to develop an innovative, comprehensive, and explainable AI vision-language foundation model (VLM) to advance the diagnosis and interpretation of brain diseases using multi-modal data. We will include patient demographics, medical imaging data (such as MRI, CT, and PET scans), histopathological data, genomic data when available, and other necessary laboratory examinations and tests to establish a screening and diagnostic model for brain diseases.

Secondary Objective: To establish a comprehensive diagnostic model with uncertainty quantification and automated report generation that covers all brain diseases based on clinical indicators. Exploratory Objective: To include MRI scans from large-scale populations for model validation.

Study Type

OBSERVATIONAL

Enrollment

100,000

Conditions

Brain (Nervous System) Cancers Brain Arterial Disease Neuro-Degenerative Disease Brain Tumors Brain Diseases Neurological di

Eligibility

Sex: ALLHealthy volunteers:

Medical Language ↔ Plain English

Inclusion Criteria: Patients with brain diseases: * Patients with brain tumors were pathologically diagnosed. * Patients with other brain diseases were correctly diagnosed. * The clinical case data of all patients were complete. Non-brain disease population: * All patients have complete clinical case data, complete brain MRI, no history brain diseases, no brain surgery or other brain diseases that affect the diagnosis and observation of MR imaging. Exclusion Criteria: * Cases in which MRI were incomplete or with significant noise and artifacts.

Outcomes

Primary Outcomes

Brain Disease Diagnostic performance

This study will evaluate how accurately the AI model can identify and differentiate between: 1. Brain tumors including gliomas, glioneuronal tumors, and neuronal tumor, meningioma, germ cell tumors, embryonal tumors, tumors of the sellar region, pineal region tumors, mesenchymal, non-meningothelial tumors, choroid plexus tumors, hematolymphoid tumors, cranial and paraspinal nerve tumors, melanocytic tumors and brain metastases based on WHO CNS 5 classification; 2. Brain diseases apart from brain tumors such as brain arterial disease, neurodegenerative disorders, etc.; 3. Normal brain findings; The model's performance will be assessed using sensitivity, specificity, F1-score AUC-ROC. Diagnostic ability of AI model will be compared against with pathological diagnosis(if possible), final clinical diagnoses by neurologists or radiologists.

Time frame: Perioperative

A Vision-Language Foundation Model for Brain Disease Diagnosis From Multimodal Data

Overview

Conditions

Interventions

Eligibility

Locations (1)

Outcomes

Primary Outcomes

Central Contacts