Building AI-Ready Datasets for Dural-Based Pathologies: A Systematic Approach to Data Curation, Annotation...

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Informatics

Background:

• Artificial Intelligence (AI) integration in neurosurgery promises enhanced diagnostics and surgical precision.
• High-quality, standardized datasets are crucial for AI development but are scarce, especially in under-resourced settings.

Purpose of the Study:

• Identify key challenges in creating AI-ready datasets for dural-based lesions.
• Propose practical solutions to overcome these barriers in dataset development.

Main Methods:

• Utilized histopathology slides from 122 patients across multiple institutions over 1 year.
• Data underwent anonymization, curation, and annotation by a multidisciplinary team.
• Implemented standardized imaging protocols, AI-assisted annotation, automated quality control, and federated learning.

Main Results:

• Encountered challenges: imaging variability, data gaps, manual annotation labor, and interobserver inconsistencies.
• Addressed data security and privacy concerns through secure transfer protocols and deidentification.
• Developed solutions including standardized protocols, AI-assisted annotation, automated QC, and federated learning.

Conclusions:

• Established a structured, high-quality dataset for AI applications in neurosurgery.
• This dataset will facilitate robust AI model development for improved diagnostic and therapeutic decision-making.
• Contributes to advancing AI-driven healthcare solutions in neurosurgery, particularly in India.