Unstructured medical image query using big data - An epilepsy case study

Area of Science:

• Medical Informatics
• Big Data Analytics
• Computational Neuroscience

Background:

• Big data technologies are crucial in medicine, necessitating new frameworks for effective data utilization.
• Querying large volumes of unstructured medical data presents challenges for hypothesis testing and patient care.

Purpose of the Study:

• To implement and assess a framework for efficient, versatile querying of unstructured medical data.
• To specifically examine the framework's feasibility within the epilepsy research domain.

Main Methods:

• A two-phase query evaluation: structured data filtering followed by distributed feature extraction on unstructured data using Hadoop.
• Development of three feature extraction modules: volume comparer, surface to volume conversion, and average intensity.
• Exploration of Hadoop configurations and comparison with a Single Server Architecture (SSA).

Main Results:

• The framework accurately executed advanced medical queries spanning structured and unstructured data.
• The surface to volume conversion module was up to 40x faster, and the average intensity module up to 85x faster on a Hadoop cluster compared to SSA.
• A 40-node Hadoop cluster processed 10,000 models in 3 hours, a task impractical for SSA.

Conclusions:

• The proposed framework demonstrates feasibility for efficient and versatile unstructured medical data querying.
• The system offers significant performance improvements over traditional architectures, advancing data-driven medicine.
• Potential exists for extending the framework's application beyond epilepsy with further research and module development.