MIST: Medical Image Streaming Toolkit

A unified framework for intelligent, progressive, and resource-efficient medical image streaming.

MIST Architecture Overview

MIST and ISLE represent two complementary frameworks addressing the challenges of large-scale medical imaging datasets and AI-driven inference systems.

Overview

MIST: Medical Imaging Streaming Toolkit

  • Challenge: Large-scale imaging datasets require significant storage and bandwidth, limiting accessibility for AI research and clinical deployment.
  • MIST Solution: A format-agnostic database enabling streaming of medical images at multiple resolutions and formats from a single high-resolution copy.
  • Evaluation: Tested across eight diverse datasets (CT, MRI, X-ray) covering multiple modalities and file formats.
  • Results: Reduced storage and bandwidth requirements without impacting image quality or downstream deep learning performance.
  • Impact: Creates a data-efficient, format-agnostic platform that reduces barriers to AI research in medical imaging.

ISLE: Intelligent Streaming for AI Inference

  • Motivation: Growing adoption of AI systems in radiology is increasing demands for bandwidth and computational resources.
  • ISLE Framework: An intelligent streaming method inspired by video-on-demand platforms to deliver only the resolution needed for AI inference using progressive encoding.
  • Results (Classification): Reduced transmission by ≥90% and decoding time by ≥87%
  • Results (Segmentation): Reduced transmission by ≥77% and decoding time by ≥89%
  • Performance: No impact on diagnostic performance (all P > 0.05).
  • Impact: Improves data and computational efficiency for AI deployment in clinical environments without compromising diagnostic accuracy.

Open-Source Tools

ComponentDescriptionRepository
MISTCore streaming and dataset management frameworkGitHub
IntelligentStreamingAI-aware streaming for real-time inferenceGitHub
OpenJPHpyPython interface for HTJ2K codecGitHub

Patents

Patent: WO2024233969A1Systems and methods for high-throughput analysis for graphical data

Filed by: University of Maryland Baltimore
Inventors: Vishwa S. Parekh, Pranav Kulkarni, Adway Kanhere, Paul H. Yi, Eliot L. Siegel

Publications

Kulkarni P., Kanhere A., Siegel E.L., Yi P.H., Parekh V.S. ISLE: An Intelligent Streaming Framework for High-Throughput AI Inference in Medical Imaging. Journal of Imaging Informatics in Medicine. 2024 Dec;37(6):3250-63. DOI

Kulkarni P., Kanhere A., Siegel E., Yi P., Parekh V.S. Towards Resource-Efficient Streaming of Large-Scale Medical Image Datasets for Deep Learning. Medical Imaging with Deep Learning (MIDL) (2025). OpenReview

Contact

Dr. Vishwa S. Parekh
UTHealth Houston
vishwa.s.parekh@uth.tmc.edu

© 2025 BioIntelligence Lab · UTHealth Houston