2023 Junior Fellow Symposium:
Shark Tank
08 June 2023 at 08:15 EDT
Room: 716A/B
The Junior Fellow Symposium presents the 3rd edition of the Shark Tank!
Subin Erattakulangara presents:
At Bridge, we are working to create a community where AI models can be shared freely with anyone who wants to use them, regardless of their technical expertise. Our open-source platform paired with user-interface building library will enable anyone who wishes to implement AI in their daily life, research, and business to do so with drag-and-drop simplicity. We realized that though AI is ubiquitous, most people are unable to apply it due to the skills barrier that exists. To implement AI models, users need to have a baseline-level of coding knowledge. This prevents many from ever being able to access AI unless they consult someone with the needed expertise, which is often costly. Our solution to this problem is an open-source platform where developers of AI models can publish their models, and users can implement these models completely online without needing to write any code.
Coaches:
Ari Borthakur, Ph.D., M.B.A., CAMIPM, University of Pennsylvania
Min Lang, M.D., Massachusetts General Hospital/Harvard Medical School
Agah Karakuzu presents:
To develop pulse sequences, MRI researchers often have to deal with arcane proprietary development libraries that have a limited selection of modules. This not only costs valuable time, but also hampers interoperable and reproducible methods development. Fortunately, vendor-neutral solutions such as pulseq, gammaSTAR, and VENUS can tackle this issue to foster efficiency and interoperability. Nevertheless, for MRI scanners to become a sandbox for research and development, there is a need for consensus vendor-neutral implementations and this is what MRI FAIR aims to provide. Similar to how Google Play and App Store can provide the same software for Android and iPhone devices, the idea for MRI FAIR is to distribute open-source pulse sequences that can run on scanners from all major vendors. This has important implications such as better inter-vendor agreement for quantitative imaging (https://qmrlab.org/VENUS) or MRS (https://www.frontiersin.org/articles/10.3389/fneur.2022.1045678/full), and for bringing AI to the clinic in a responsible manner (https://insightsimaging.springeropen.com/articles/10.1186/s13244-020-00931-1).
Coaches:
Sairam Geethanath, Ph.D., Mt. Sinai School of Medicine
Tatiana Wolfe, Ph.D., University of Arkansas for Medical Sciences
Wolfgang Loew presents:
This platform enables the use of anatomically specific multinuclear phased arrays in combination with a volume birdcage coil. This setup allows for the operation of a birdcage as a transmit/receive (T/R) coil or with the birdcage as a transmit-only coil and the phased array as receiver. Xenon-129 at 3 Tesla was chosen as an example. Such a dedicated platform is not limited to a single nucleus and could easily be adapted to the frequency of another nucleus at any field strength.
Coaches:
Martijn A. Cloos, Ph.D., University of Queensland
Linda Knutsson, Ph.D., Kennedy Krieger Institute
Paulina Šiurytė presents:
The acoustic noise in MRI is a major source of patient discomfort and greatly affects patient acceptance of MRI scans. We have developed a new system, predictive noise canceling, which predicts the MRI sounds directly from the sequence gradients. A playback through an MR compatible headphone system as anti-noise creates a silent zone for the patient and enables a cost-effective solution for MR scanning with improved patient comfort.
Coaches:
Aleksandar N. Nacev, Ph.D., Promaxo
Ante Zhu, Ph.D., GE Global Research
Suyash Mohan presents:
Problem Statement: Given the explosive growth of radiology exams in the last two decades, and that the number of radiologists trained every year has stayed relatively flat, it is projected that in the next 10 years we will have a shortage of over 11,000 radiologists in the US, which will be even greater in resource poor countries. In order for us to bridge this gap, we will need “intelligent AI tools” to assist in medical image interpretation.
Market Gap: As of now, there are 51 FDA-approved AI devices for brain MRI and CT image analysis, but they only tackle two broad categories of findings: brain volumetrics and stroke. This is highly insufficient as there is a wide variety of neurological diseases that are not addressed by any existing AI tool/platform.
Market Size: On an average, 110 million CT & MRI scans are performed annually in the USA, with the projected number about 500 million for CT & MRI worldwide. There are approximately 35,000 radiologists in the U.S. with approximately $20 billion charged for professional services and $80 billion charged for technical services, making radiology a $100 billion market!
Technical Solution: We have developed a powerful image analysis tool that provides a radiological diagnosis and a full draft report by combining the power of state-of-the-art neural nets and advanced Bayesnets. Our startup medical software company, GalileoCDS, uses contemporary AI/ML methods to analyze and report radiological images. It is based on patent-protected IP owned by the University of Pennsylvania, which is licensed to Galileo. Our introductory product, Galileo Consultant, uses highly refined CNNs to extract and report the important radiographic features from brain MRI scans and a unique Bayesnet to infer differential diagnosis based on these key features. The product is designed to function as an intelligent radiological assistant, providing a draft report ready for final review and approval by the radiologist. The product is carefully integrated into clinical workflow and can improve a radiologist’s efficiency up to 30%. Research trials have shown the diagnostic accuracy of Galileo Consultant to be comparable to that of neuroradiology fellows, who also create draft reports, 80% of which are signed without change by attending radiologists. The initial clinical version of this Galileo Consultant is now being installed under research conditions and IRB approval at the University of Pennsylvania.
GalileoCDS also offers an educational version of the product, Galileo Education, which is designed for medical students, residents, fellows, non-specialty-trained radiologists, and other healthcare workers. This product is a non-clinical teaching tool that is now available on the web. Not only is Galileo Education an effective, standalone, educational tool, but it also serves as an introduction to our clinical product.
Competitive Landscape: Existing image analysis software applications only cover one or two neurological conditions. Ours is the only product that offers a complete library of all 124 adult brain diseases. We have run pilot studies on hundreds of cases and have documented that the diagnostic accuracy of our software is equal to that of an “ABR (American Board of Radiology)-certified neuroradiology fellow.” Our software not only works for common diseases, but it can also diagnose rare diseases. It is transformative, as it creates a real report that is indistinguishable from that of a human’s. The created draft report is accurate, ready to be signed, saves significant time for the radiologist, and covers the entire spectrum of brain diseases.
Team: Our vision and an exceptional team with over 100 plus years of combined neuroradiology experience and Ph.D. computer scientists is our strength. We have engineered our product in such a way that there is no additional effort for the radiologist; every output is completely and directly integrated into the clinical workflow. What is unique about our approach is that this hybrid model generates a differential diagnosis that is “explainable” and writes a full draft report combining domain knowledge with computer vision mimicking the way radiologists practice.
Coaches:
José Pedro Marques, Ph.D., Radboud University
Mustapha Bouhrara, Ph.D., NIH/NIA