Luca Giancardo is an Associate Professor at the Center for Precision Health within McWilliams School of Biomedical Informaticsand UTHealth Houston. He has co-appointments at the McGovern Medical School and the Institute for Stroke and Cerebrovascular Diseases at UTHealth Houston. He directs the Giancardo Lab.
He is a computer scientist with extensive experience in image analysis and machine learning. He has worked on developing new machine learning-based methodologies to discover computational biomarkers from patterns in biomedical data such as optical images, magnetic resonance imaging, X-rays, computer tomography, laboratory animal videos or tracking devices. His work has been applied to a number of biomedical applications, such as stroke diagnosis, diabetic retinopathy screening or neurodegenerative disease tracking and successfully translated to industry with two startups based on his methods. He has authored/co-authored more than 70 peer-reviewed articles which were featured by news outlets such as MIT Technology Review, Smithsonian magazine, and others. He has received multiple awards, including the prestigious 100k Singapore Challenge (judging panel composed of Nobel Prize and Millennium Technology Prize winners). One of his projects (neuroQWERTY) has been included in the MIT Museum permanent exhibit “Essential MIT”.
He was awarded competitive grants from NIH (R01), Translational Research Institute for Space Health, foundations, and private companies.
- Tell us about your research center and/or what research/work you are currently working on?
Main projects
Deep Learning Enabled Endovascular Stroke Therapy Screening in Community Hospitals
The goal of this project is to develop and validate machine learning-based software tools to identify key brain imaging measurements in acute stroke patients using brain imaging modalities available in the majority of the hospitals such as CT angiography. This would enable better management of stroke patients and better delivery of endovascular stroke therapy
Actionable Stroke Detection with Deep Learning and Retinal Imaging
This project aims to develop systems and data for identifying acute strokes using retina imaging. This would enable the deployment of life saving drugs in situations where brain scans are typically not available such as ambulances or even space missions.
Endometriosis Screening Tool
Develop automated algorithms to screen for endometriosis signs from MRI images
Stroke care from Bench to Bedside
We are integrating our machine learning algorithms for stroke care into the acute stroke care pathway in a manner that sends automated alerts of its findings to physicians on the care team. This allows us to evaluate their real world impact.
Neurodegeneration Markers with Digital Device Interaction and Machine Learning
We are developing new computational methods to detect subtle motor impairments based on timing of finger interactions on a standard personal computer or mobile devices without predefined tests. We created and tested software tools to remotely collect months’ worth of data and computational algorithms able to repeatably detect signs of neurodegenerative diseases.
Machine Learning-based Tools for Longitudinal Brain Imaging Analysis and Connectomes
We develop techniques to analyze the changes in structural connectivity and voxel-based changes of the brain longitudinally. These techniques allow for data-driven processing able to discover imaging biomarkers candidates from whole brain or atlas specific diffusion-MRI data. These have application for neurodegenerative diseases such as Parkinson's Disease or Alzheimer. We have also developed approaches for combining imaging and clinical data and sparse machine learning-based method to automatically explore inter-group differences in Diffusion-MRI connectomes in an unbiased-fashion.
Machine Learning-based Tools for Retina Image Analysis
General purpose tools for retina image analysis and imaging biomarkers extraction using machine learning-based models with fundus images OCT and OCT-A images
Deep Learning-based Tools for MRI Safety
MRI is an imaging modality essential for the diagnosis, prognosis and tracking of several medical condition. However, people with implantable devices are likely to lack accurate information about the specifics of their device. Systems to allow for a quick and safe MRI safety clearance from X-ray images could be tightly integrated with the current radiologist workflow and leverage modern machine learning (ML) algorithms able to identify and quantify image patterns and being continuously updated with newer devices entering the market.
- What type of student or Postdoctoral Fellow are you looking for to work in your center?
Student with interest in developing machine learning methods for medical image (or signal) analysis. We are looking for candidates with computational background, for method development, and more medical/clinical background for data collection and labeling.
- What does the future of your research look like?
Development of new foundational machine learning models that allow generalizability with minimal training data. This will be enabled also new cutting edge methods using the integration of imaging and text data and enabled by the data at Memorial Herman containing millions of imaging studies.
- What does the future of informatics look like?
The future of informatics is expected to be characterized by the integration of machine learning in all aspects of healthcare domains, so much so that these techniques will follow a commoditization process and taken for granted in any informatics application. In a similar way as databases are right now. Precision medicine will thrive through the amalgamation of genomic, imaging and patient records. Emphasis will be on data security, interdisciplinary collaboration, and the transformation of healthcare and environmental management. Human-computer interaction will become more intuitive, education will adapt, and ethical considerations surrounding data ownership and unbiased and interpretable ML deployment will gain prominence.
- What courses do you teach?
BMI 6331 Medical Imaging and Signal Pattern Recognition: Biomedical data in the form of images, videos or other unstructured signals are continuously collected by clinicians, such as radiologists, dermatologists or ophthalmologists, life science researchers and increasingly by ourselves with our personal devices. Tools able to distill quantitative actionable information from these data are essential to generate phenotypes, aid diagnosis, screening, treatment and automate repetitive tasks. In the era of personalized medicine and big data, they have become an urgent medical need. In this course, you will be introduced to the essential pattern recognitions techniques to build and evaluate such tools. We will be covering the basics of image/signal processing, computer vision and applied machine learning (including deep learning) with hands on examples relevant to biomedical applications.
- What major UTHealth Houston departments/institutes do you collaborate with?
Neurology, McGovern Medical School
Diagnostic and Interventional Imaging, McGovern Medical School
Institute for Stroke and Cerebrovascular Diseases
