Research: Medical Image Processing Systems (MIPS) Labs

Medical Image Processing Systems (MIPS) Laboratory

Medical Images have become a significantly useful mechanism for diagnostic medicine today. Few primary challenges in medical image analysis include image denoising, information extraction and feature detection for particular pathologies. Our goal is to enhance the image signal to noise ratio, estimate noise parameters and extract features and to develop scalable algorithms to identify pathologies and abnormalities in terms of measurable bio-informatics.

Current Open Research Projects

1. Scalable Machine learning using the Cloud-computing resources in Microsoft Azure Machine Learning Studio platform.

Development of high-speed machine learning framework for medical image analysis. This project involves leveraging the hardware/device independence of the cloud-computing platform of Microsoft Azure Machine Learning Studio (MAMLS). This project is directed towards building experimental machine learning framework that can process several TB of medical image data for assessment and pathology diagnostic operations. The Microsoft Azure Machine Learning platform will be used as an educational tool to expose UWB students to the concepts of cloud-computing, machine learning and Big Data.

The MAMLS platform is used to implement deep-learning from medical images using 2D Convolutional Neural Networks (CNNs). The goal is to learn new features that will enhance medical image based diagnostic systems. Additionally, the CNN models are being extended to incorporate 3D volumetric object scans.

2. Automated CT image quality estimation in collaboration with Department of Radiology, University of Washington.

Automated image quality estimation using Computed Tomography (CT) images in collaboration with the Department of Radiology, University of Washington, Seattle. This work involves estimation of novel and automated image based informatics that can be used for generating diagnostic quality images under low contrast enhancement dosages. The goal of this research is to develop automated modules that correlate with the manual understanding of image quality to allow enhanced visualization and diagnostic protocol enhancement [A Seed grant from Radiology Society of North America (RSNA) granted, and a UW RRF Grant received].

3. Automated analysis of Research topic trends in Management Sciences, in collaboration with with the School of Business, University of Washington, Bothell.

Automated Topic Analysis in Business and Management Research. This project is in collaboration with the Business School in University of Washington, Bothell, Cornell University and Ohio State University. The goal of this project is to identify research topics in the past 60 years that have dictated changes in the realm of Business and Management strategies. Several network-based mathematical models have been developed so far for the underlying pattern recognition in scholarly material topic trends. These mathematical models will be extended to quantitative analysis of supplier evaluation using supplier 10K data records.

4. Automated Facial expression detection methods.

Automated facial expression detection. This work involves the use of complex networks-based modeling for facial expressions and network partition/clustering for high reliability and low computational complexity for expression recognition tasks. A survey paper regarding facial data bases and quantitative models developed so far has been published.

5. Automated "Anemia-like Paleness" detection using smart-phones.

As a follow-up on automated facial expression detection, I have been working on development of Smart health monitoring systems using “paleness index” that is measured using facial images. The combination of smart-phone based images with an online decision making algorithm that detects the severity of paleness will significantly increase the awareness of women and children that are currently at risk of anemia, but have limited access to healthcare on a regular basis. Such a point-of care diagnostic design will bring healthcare to the fingertips of the public and increase the resourcefulness of the present day health-care delivery system. Preliminary results on a set of uncontrolled test-bed images are supportive of automated screening application development.

6. Automated detection of hypertensive retinopathy and proliferative diabetic retinopathy. Collaboration with University of Minnesota.

Automated detection of Hypertensive Retinopathy (HR) in correlation with risk of stroke and organ damage. This work is in collaboration with the School of Medicine at University of Minnesota. Preliminary work has shown that blood vessel classification and optimal feature identification can be useful for enhancing the accuracy of the estimation of arterio-venous ratio and the severity of hypertensive retinopathy using fundus images.

This work is based on spatial and frequency-based filters to segment retinal fundus images followed by invoking machine learning algorithms for classification of images with varying degrees of pathology.

7. Smart-wearable electronic devices for health monitoring.

Development of a commercial wearable device for brain wave monitoring.

Research Highlights!