Cloud Engineering Principles and Technology Enablers for Medical Image Processing-as-a-Service.

Area of Science:

• Cloud Computing
• Medical Imaging Informatics
• Big Data Technologies

Background:

• Traditional medical imaging storage (NFS, COINS, XNAT) faces network bandwidth limitations.
• Cloud-based solutions like Apache Hadoop and HBase offer scalable storage and computation.
• HBase's default data distribution hinders hierarchical medical imaging data organization.

Purpose of the Study:

• To propose cloud engineering principles and technology enhancements for medical imaging on Hadoop/HBase.
• To address HBase's limitations in handling hierarchical medical imaging data.
• To improve the efficiency of medical image processing in a cloud environment.

Main Methods:

• Designed a hierarchical row-key structure for HBase tailored to imaging data.
• Developed a novel data allocation policy in HBase to enforce data collocation.
• Integrated these enhancements into the Apache Hadoop ecosystem for medical imaging.

Main Results:

• Achieved a three-fold speed improvement in DICOM to NiFTI conversion compared to default HBase.
• Demonstrated a nearly six-fold improvement over Network File System (NFS) approaches.
• Reduced file access latency compared to traditional network-attached storage.

Conclusions:

• The proposed HBase enhancements optimize cloud-based medical image processing.
• The approach maintains compatibility with existing analysis procedures and software.
• This work accelerates data processing by minimizing network usage and localizing computation.