Rakesh Shiradkar, PhD is Research Assistant Professor at Department of Biomedical Engineering, Case Western Reserve University, Cleveland OH, USA. His research focus includes building novel computational tools and machine learning models for characterization, diagnosis, and prognosis of cancer on imaging.
Dr. Rakesh talks to Biopatrika about his interest for personalized diagnostics and its applications in Healthcare.
What attracted you to Personalized Diagnostics?
I was trained in Electrical Engineering and Computer Science through my undergraduate at the Indian Institute of Technology Roorkee and graduate studies at the National University of Singapore. This was exciting; however, I felt a bit of disconnect when I could not tangibly see something directly impacting humanity in their daily lives. When I joined Case Western Reserve University and saw the close-knit ecosystem between engineering, medical school and health care centers, cutting edge research in AI assisted diagnostics, I was immediately drawn to it and wanted to make an impact in this field. Several factors affect patient diagnosis and how they respond to therapies and a one-size-fits-all is rarely a solution. When I became aware of the amount of information that exists within patient records, imaging and follow up data, I realized so much of this information is not being effectively utilized. With the advancement of imaging technologies, computational power, sophisticated AI-based technologies, I recognized that there is immense potential to integrate all these variables to design personalized diagnostic tools that can help clinicians make better-informed decisions.
What are your ongoing projects?
Currently, I am developing predictive models for prostate cancer patients to identify which patients need definitive treatment and those that do not. Especially in the United States, a significant overdiagnosis and overtreatment are going on in men diagnosed with prostate cancer. One of the main reasons for this is invasive diagnostic methods such as biopsies and blood tests, causing significant side effects, including anxiety, side effects and missing cancer. Non-invasive, imaging-based diagnosis with MRI is still limited by resolution, confounders, and variations in diagnostic reads. Image processing with advanced computation power has the potential to tease out sub-visual signatures from medical imaging modalities that can characterize underlying disease patterns. We are developing risk-stratification tools with AI methods to accurately localize cancer on MRI and predict with low-risk disease without immediate treatment.
I am also working on building prognostic models for cancer patients to predict who would experience adverse outcomes such as disease recurrence and metastasis following radiotherapy or surgery. Several cancer patients experience recurrence of disease due to left-over cancer cells or poor response to therapy. Pre-emptively identifying such patients allows for administering adjuvant and neo-adjuvant therapies. I am working on integrating imaging, pathology, clinical and demographic factors together with AI-based methods to develop comprehensive nomograms that can help clinicians identify which patients would favorably respond to therapy. Taking a multi-modal approach can result in designing comprehensive predictive models considering several data streams to arrive at a decision.
Other projects include identifying adverse pathology, minimizing health disparities, and quantifying treatment-related changes in cancer patients.
“[…] developing risk-stratification tools with AI methods to accurately localize cancer on MRI and predict those that have a low-risk disease without the need for immediate treatment.”
What are your future goals?
Currently, most AI-based studies in healthcare are on retrospectively acquired data. For these to be used in real-time at hospitals, we need large scale validation of these methods. My future endeavors would be to work closely with clinicians and design large scale clinical trials to validate the technologies developed in the lab. A part of this will involve acquiring federal funds to support these trials. This will ensure that these methods are not limited to research alone and can benefit patients. I also wish to expand to other disease sites from the prostate to kidney diseases, liver, pancreatic cancers. This will largely depend on building strong collaborations with clinicians and health care centers. One of my goals is also to benefit low- and middle-income countries. Growing up in India, I realize that there is a significant shortage of clinicians per capita and AI technologies can play a major role in lessening the burden of healthcare workers. I plan to collaborate with major health care centers in India to bring these technologies into their day to day workforce.
What is your vision to translate basic science to products?
Translation of lab-developed science and technology to usable products is crucial for these technologies to truly benefit population at large. Following large-scale clinical trials validation, I believe science will be ready to pass the FDA regulatory within the United States. We have recently received pilot grant funding from the Clinical and Translational Science Collaborative to bring our technology for low-risk prostate cancer monitoring closer to clinical deployment. We have also conducted extensive interviews with clinicians, tech personnel and regulatory experts, which taught us a great deal about what it takes for the technology to be converted into a commercially viable product. There is a strong value proposition for our technology, however, this must be appropriately tailored to different markets. The focus in developed and strong economies will be to provide added confidence to clinicians in making decisions. In contrast, in low and middle economies, where clinicians are overburdened by a large volume of workload, these technologies would play a supporting role in minimizing the burden.
How your work will contribute to society?
The standard of care clinical exams performed today are invasive, mono-modal, and expensive. Biopsy procedures are invasive, painful, cause side effects and occasionally miss out on cancer focus. Predictive models using genomics are expensive — for example, Oncotype DX is a genomic test that predicts the likelihood of cancer recurrence (shown to be effective for breast cancer) and it costs about $4000. Similarly, Decipher TMis another genomic test that costs around the same, predicting the likelihood of metastasis in prostate cancer. On the other hand, using imaging-based, AI-assisted test uses routinely acquired scans and demographic information to predict the same and performs comparably or even better, based on our preliminary data. And this is tissue non-destructive, can be securely performed anywhere in the world with access to computational resources allowing non-invasive, accurate and personalized diagnostics at a fraction of cost.
Originally published at http://biopatrika.com on January 6, 2021.