1333-GRP HIMLT: A Hybrid Interactive Machine Learning Tool for Whole Slide Image Analysis

Presenter Information

Location

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Document Type

Event

Start Date

20-11-2026 2:00 PM

Description

Whole Slide Images (WSIs) are high-resolution dig ital pathology images created for representing the full tissue specimen by scanning entire glass microscope slides. Tradition ally, pathologists have manually examined these slides under a microscope, a highly time-consuming process because a single WSI contains billions of pixels, which has led human experts to struggle with WSI analysis, even causing inter-intra observer variability. In this paper, we present a new hybrid interactive machine learning tool (HIMLT) for WSI analysis that leverages feature fusion delivering an enterprise-grade user experience for cancer detection. Our tool has been developed based on Hybrid Modality Attention Fusion (HMAF) from our earlier work by projecting both feature types into a shared latent space. HIMLT adopts a well-known machine learning algorithm, Random Forest, for faster iterative training and features a modular design with RESTful APIs to enable extensibility and collaborative workflows. Our work has been distributed as a Docker application with five containerized services including Ng inx, Flask, MySQL, Redis, and IIPImage, enabling reproducible, single-command deployment on any hardware without manual configuration, advancing digital pathology through feature fusion and user-friendly interactive design. To bridge computational predictions and clinical relevance, a Kaplan-Meier survival analysis module was integrated into HIMLT. Using TCGA GBM patient data, the proportion of malignant nuclei predicted per whole-slide image serves as a prognostic biomarker to stratify patients into risk groups. Moreover, five conversational LLM assistants supporting both cloud-based and locally deployed language models were added for HIMLT to allow users to query the system.

This document is currently not available here.

Share

COinS
 
Nov 20th, 2:00 PM

1333-GRP HIMLT: A Hybrid Interactive Machine Learning Tool for Whole Slide Image Analysis

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Whole Slide Images (WSIs) are high-resolution dig ital pathology images created for representing the full tissue specimen by scanning entire glass microscope slides. Tradition ally, pathologists have manually examined these slides under a microscope, a highly time-consuming process because a single WSI contains billions of pixels, which has led human experts to struggle with WSI analysis, even causing inter-intra observer variability. In this paper, we present a new hybrid interactive machine learning tool (HIMLT) for WSI analysis that leverages feature fusion delivering an enterprise-grade user experience for cancer detection. Our tool has been developed based on Hybrid Modality Attention Fusion (HMAF) from our earlier work by projecting both feature types into a shared latent space. HIMLT adopts a well-known machine learning algorithm, Random Forest, for faster iterative training and features a modular design with RESTful APIs to enable extensibility and collaborative workflows. Our work has been distributed as a Docker application with five containerized services including Ng inx, Flask, MySQL, Redis, and IIPImage, enabling reproducible, single-command deployment on any hardware without manual configuration, advancing digital pathology through feature fusion and user-friendly interactive design. To bridge computational predictions and clinical relevance, a Kaplan-Meier survival analysis module was integrated into HIMLT. Using TCGA GBM patient data, the proportion of malignant nuclei predicted per whole-slide image serves as a prognostic biomarker to stratify patients into risk groups. Moreover, five conversational LLM assistants supporting both cloud-based and locally deployed language models were added for HIMLT to allow users to query the system.