A Budding Computer Scientist, Shreyas Fadnavis Making Substantial Contributions in the Field of Medical Imaging

A budding Computer Scientist, Shreyas Fadnavis from Harvard University, Ph.D from Indiana University Bloomington making substantial contributions to the field of machine learning for medical imaging. His important career goal is to revolutionize the healthcare industry particularly in the area of medical imaging diagnosis.

What draws you to the field of computing, particularly in machine learning?

Significantly inspired by Richard Feynman’s ‘The Pleasure of Finding Things Out’ and George Pólya’s ‘How to Solve It.’ These seminal works instilled in me a profound curiosity and a systematic approach to problem-solving from a young age. My undergraduate studies at the Pune Institute of Computer Technology, University of Pune, where I pursued a Bachelor of Engineering in Computer Engineering, laid the foundation for my interest in computing. Courses like Data Structures, Algorithms, Natural Language Processing, and Artificial Intelligence were not just academic pursuits; they sparked a curiosity in me about the vast potential of computing.

What particularly draws me to the field of computing, and more specifically to machine learning, is the blend of rigorous mathematical theory and practical, impactful applications. My graduate studies at Indiana University Bloomington, where I earned both my Master’s and PhD in Engineering with a focus on Machine Learning, further cemented this interest. Here, I delved into Bayesian Theory, Image Processing, Combinatorial Topology and Self-supervised Learning, gaining a deep appreciation for the power of machine learning in interpreting and analyzing complex data.

My professional experiences have also significantly shaped my passion. Working with cutting-edge projects at Johnson & Johnson, Harvard, and IBM Research, I’ve been able to apply machine learning techniques to real-world problems in biomedical engineering and neuroimaging. Developing self-supervised architectures for medical image analysis and contributing to multimodal machine learning projects have been particularly fulfilling, allowing me to see the tangible impact of my work.

In summary, my path towards becoming a computer scientist has been fueled by a combination of academic rigor, practical application.

What was your first top-tier project?

Reflecting on my array of projects, I believe the most impactful has been my work on “Patch2Self,” a novel self-supervised learning framework for denoising diffusion MRI data. This project stands out due to its innovative approach and significant implications in the field of medical imaging.

Developed during my tenure as a PhD candidate and a core team member with the Diffusion Imaging in Python (DIPY) project, Patch2Self represents a major advancement in the processing of diffusion MRI data. It’s an achievement not just in terms of technical complexity but also for its potential to enhance the accuracy and reliability of medical diagnoses.

The impact of Patch2Self is manifold. Firstly, it addresses a critical challenge in neuroimaging - the denoising of diffusion MRI data, which is crucial for accurate imaging and subsequent analysis. By leveraging self-supervised learning, Patch2Self improves the quality of imaging data without the need for external training data, making it a versatile and robust solution.

Additionally, Patch2Self’s success was recognized with a Spotlight Presentation at the prestigious Neural Information Processing Systems (NeurIPS) conference in 2020, marking it as a top-tier contribution in the field.

Which type of machine learning techniques have you used in your projects related to medical imaging?

In my various projects related to medical imaging, I have employed a range of sophisticated machine learning techniques tailored to the specific challenges and complexities of the field.

One key technique I’ve used is Self-Supervised Learning, prominently featured in the development of Patch2Self for denoising diffusion MRI data. This method allowed for effective training of models directly from the unlabeled data, eliminating the need for externally labeled training datasets. This approach was crucial in enhancing the quality and reliability of medical imaging data, particularly in neuroimaging.

Another significant technique in my repertoire is Multi-View Learning, which I developed during my tenure as a PhD Research Intern at IBM Research. This framework was instrumental for the intermediate data fusion of diffusion MRI microstructure models.

Deep Learning has been a cornerstone in many of my projects. For instance, at Johnson & Johnson, I developed a deep learning-based self-supervised architecture for detecting Pulmonary Hypertension from echocardiography images.

During my Master’s studies, I focused on Bayesian Theory, a fundamental approach for probabilistic modeling in medical imaging. Bayesian methods are particularly beneficial in handling uncertainties and incorporating prior knowledge, which is crucial in medical diagnosis and research.

Topological Optimization was another key area of my research, especially in projects like Intravoxel Incoherent Motion using Topological Optimization and Variable Projection.

In addition to these, my work has also involved Random Matrix Algorithms, which are critical for signal recovery in medical imaging.

Lastly, my involvement in projects such as NLP-Transformers and Explainable AI at Harvard Medical School & IBM Research, where I worked on detecting and summarizing sub-types of Chronic Pain and Psychosis, showcases the breadth of my expertise.

How do you plan to contribute to contributing significantly in the field of machine learning for medical imaging in the future?

Firstly, I plan to continue innovating in algorithm development, building upon my work with projects like Patch2Self and my involvement with the Diffusion Imaging in Python (DIPY) project. My focus will be on creating advanced machine learning algorithms that enhance image quality, improve diagnostic accuracy, and extract meaningful insights from medical imaging data.

In my ongoing roles at J&J, I have the opportunity to delve deeper into areas such as multimodal machine learning, time-series modeling, and interpretable machine learning. I aim to use these platforms to develop more sophisticated models capable of handling the complexities and variabilities inherent in biomedical data.

Self-supervised learning techniques, which have been successful in my previous projects, will also be a key area of focus. I intend to further explore and refine these techniques to develop models that can learn effectively from large, unlabeled datasets, which are often encountered in medical imaging.

Interdisciplinary collaboration is another cornerstone of my future contributions. Finally, I am excited to explore new challenges that are emerging in medical imaging.

What is your most important career goal as a budding computer scientist? Are you working on any innovative tech solution for transformation of healthcare industry in diagnosis?

As a budding computer scientist, my most important career goal is to revolutionize the healthcare industry, particularly in the area of medical imaging diagnosis. My focus is on harnessing the potential of Machine Learning (ML) and Artificial Intelligence (AI) to develop innovative technologies that can significantly improve the accuracy and efficiency of medical diagnoses.

Currently, I am engaged in cutting-edge projects that exemplify this ambition. At Johnson & Johnson, I have worked on self-supervised learning for different applications of medical imaging such as echocardiography and endoscopy. This project, at the intersection of computer vision and medical imaging, represents a significant stride in using advanced ML techniques for more precise and automated imaging analysis.

In addition to these roles, my involvement with the Diffusion Imaging in Python (DIPY) project as a core-contributor showcases my dedication to advancing medical imaging technology. Through DIPY, I have contributed to the development of algorithms for denoising diffusion MRI data, improving the clarity and reliability of neuroimaging.

My goal is to continue exploring and innovating in the field of medical imaging, using ML to push the boundaries of what’s possible in diagnosing and understanding various medical conditions. The integration of advanced computational techniques in medical imaging is not just about technological advancement; it’s about creating tools that can lead to better patient outcomes and more personalized healthcare.

In essence, my career aspiration is to be at the forefront of transforming healthcare through technological innovation in medical imaging, making diagnosis faster, more accurate, and more accessible.

COMMENT NOW