Deep learning has disrupted nearly every major field of study from computer vision to genomics. The unparalleled success of these models has, in many cases, been fueled by an explosion of data. Millions of labeled images, thousands of annotated ICU admissions, and hundreds of hours of transcribed speech are common standards in the literature. Clinical neuroscience is a notable holdout to this trend. It is a field of unavoidably small datasets, massive patient variability, and complex (largely unknown) phenomena. My lab tackles these challenges across a spectrum of projects, from answering foundational neuroscientific questions to translational applications of neuroimaging data to exploratory directions for probing neural circuitry. One of our key strategies is to integrate a priori information about the brain and biology into the model design.

This talk will highlight two ongoing projects that epitomize this strategy. First, I will showcase an end-to-end deep learning framework that fuses neuroimaging, genetic, and phenotypic data, while maintaining interpretability of the extracted biomarkers. We use a learnable dropout layer to extract a sparse subset of predictive imaging features and a biologically informed deep network architecture for whole-genome analysis. Specifically, the network uses hierarchical graph convolution that mimic the organization of a well-established gene ontology to track the convergence of genetic risk across biological pathways. Second, I will present a deep-generative hybrid model for epileptic seizure detection from scalp EEG. The latent variables in this model capture the spatiotemporal spread of a seizure; they are complemented by a nonparametric likelihood based on convolutional neural networks. I will also highlight our current end-to-end extensions of this work focused on seizure onset localization. Finally, I will conclude with exciting future directions for our work across the foundational, translational, and exploratory axes.

Roughly 40% of the population will be diagnosed with some form of cancer in their lifetime. In a large majority of these cases, a definitive cancer diagnosis is only possible via histopathologic confirmation on a tissue slide. With the increasing popularity of the digitization of pathology slides, a wealth of new untapped data is now regularly being created.

Computational analysis of these routinely captured H&E slides is facilitating the creation of diagnostic tools for tasks such as disease identification and grading. Further, by identifying patterns of disease presentation across large cohorts of retrospectively analyzed patients, new insights for predicting prognosis and therapy response are possible [1,2]. Such biomarkers, derived from inexpensive histology slides, stand to improve the standard of care for all patient populations, especially where expensive genomic testing may not be readily available. Moreover, since numerous other diseases and disorders, such as oncoming clinical heart failure [3], are similarly diagnosed via pathology slides, those patients also stand to benefit from these same technological advances in the digital pathology space.

This talk will discuss our research aimed towards reaching the goal of precision medicine, wherein patients receive optimized treatment based on historical evidence. The talk discusses how the applications of deep learning in this domain are significantly improving the efficiency and robustness of these models [4]. Numerous challenges remain, though, especially in the context of quality control and annotation gathering. This talk further introduces the audience to open-source tools being developed and deployed to meet these pressing needs, including quality control (histoqc.com [5]), annotation (quickannotator.com), labeling (patchsorter.com), validation (cohortfinder.com).

This presentation will identify the gap between AI accuracy and making good clinical decisions. I will present a study where we develop an ethical framework for clinical decision-making that can help clinicians meet medicolegal and ethical standards when using AI that does not rely on explainability, nor perfect accuracy of the model.

Dr. Marzyeh Ghassemi focuses on creating and applying machine learning to understand and improve health in ways that are robust, private and fair. Dr. Ghassemi will talk about her work trying to train models that do not learn biased rules or recommendations that harm minorities or minoritized populations. The Healthy ML group tackles the many novel technical opportunities for machine learning in health, and works to make important progress with careful application to this domain.

The dream of precision health is to develop a data-driven, continuous learning system where new health information is instantly incorporated to optimize care delivery and accelerate biomedical discovery. In reality, however, the health ecosystem is plagued by overwhelming unstructured data and unscalable manual processing. Self-supervised AI such as large language models (LLMs) can supercharge structuring of biomedical data and accelerate transformation towards precision health. In this talk, I'll present our research progress on biomedical AI for precision health, spanning biomedical LLMs, multi-modal learning, and causal discovery. This enables us to extract knowledge from tens of millions of publications, structure real-world data for millions of cancer patients, and apply the extracted knowledge and real-world evidence to advancing precision oncology in deep partnerships with real-world stakeholders.

Cancer is a heterogeneous disease, with known inter-tumor and intra-tumor heterogeneity in solid tumors. Established histopathologic prognostic biomarkers generally acquired from a tumor biopsy may be limited by sampling variation. Radiomics is an emerging field with the potential to leverage the whole tumor via non-invasive sampling afforded by medical imaging to extract high throughput, quantitative features for personalized tumor characterization. Identifying imaging phenotypes via radiomics analysis and understanding their relationship with prognostic markers and patient outcomes can allow for a non-invasive assessment of tumor heterogeneity. Recent studies have shown that intrinsic radiomic phenotypes of tumor heterogeneity for cancer may have independent prognostic value when predicting disease aggressiveness and recurrence. The independent prognostic value of imaging heterogeneity phenotypes suggests that radiogenomic phenotypes can provide a non-invasive characterization of tumor heterogeneity to augment genomic assays in precision prognosis and treatment.

As the market leader in deep learning and parallel computing, NVIDIA is fully committed to advancing applied research in medical imaging. Our goal is to revolutionize the capabilities of medical doctors and radiologists by equipping them with powerful tools and applications based on deep learning. We firmly believe that the integration of deep learning and accelerated AI will have a profound impact on the life sciences, medicine, and the healthcare industry as a whole. To drive this transformative process, NVIDIA is actively democratizing deep learning through the provision of a comprehensive AI computing platform specifically designed for the healthcare community. These GPU-accelerated solutions not only promote collaboration but also prioritize the security of each institution's information. By doing so, we are fostering a collective effort in harnessing the potential of deep learning to benefit healthcare.

During this talk, I will showcase remarkable research achievements accomplished by NVIDIA's deep learning in medical imaging team. This includes breakthroughs in segmentation, self-supervised learning, federated learning, and other related areas. Additionally, I will provide insights into the exciting avenues of research that our team is currently exploring.

Generating factually correct medical text is of utmost importance due to several reasons. Firstly, patient safety is heavily dependent on accurate information as medical decisions are often made based on the information provided. Secondly, trust in AI as a reliable tool in the medical field is essential, and this trust can only be established by generating accurate and reliable medical text. Lastly, medical research also relies heavily on accurate information for meaningful results.

Recent studies have explored new approaches for generating medical text from images or findings, ranging from pretraining to Reinforcement Learning, and leveraging expert annotations. However, a potential game changer in the field is the integration of GPT models in pipelines for generating factually correct medical text for research or production purposes.

Computer science, in the form of Artificial Intelligence (A.I.), has been applied to both the MR imaging and the pathological analysis of many types of tumors within the genitourinary system. Using the prostate and the adrenal glands as examples, the role of AI in differentiating normal tissue from tumors, and in deciphering clinically significant from non-clinically significant cancers will be discussed. In addition, the future potential for A.I. to interrogate the underlying microscopic tissue architecture and how this could impact clinical care will be discussed.

In this talk, I will provide arguments for the thesis that nearly all diagnostic radiology could be performed by computers and that the notion that AI will not replace radiologists is only temporarily true. Some well-known and lesser-known examples of AI systems analyzing medical images with a stand-alone performance on par or beyond human experts will be presented. I will show that systems built by academia, in collaborative efforts, may even outperform commercially available systems. Next, I will sketch a way forward to implement automated diagnostic radiology and argue that  this is needed to keep healthcare affordable in societies wrestling with aging populations. Some pitfalls, like excessive demands for trials, will be discussed. The key to success is to convince radiologists to take the lead in this process. They need to collaborate with AI developers, but AI developers and the medical device industry should not lead this process. Radiologists should, in fact, stop training radiologists, and instead, start training machines.

We evaluated the ability to detect causal associations among diseases and imaging findings from radiology reports. We analyzed 1,702,462 consecutive reports of 1,396,293 patients for positive mention of 16,839 entities (disorders and imaging findings) of the Radiology Gamuts Ontology (RGO).  The approach, which combined NLP techniques to identify relevant terms and Bayesian networks to infer causality identified 725 pairs of entities as causally related; 634 were confirmed by reference to RGO or physician review (87% precision).  This approach finds causal relationships with high precision from textual radiology reports, despite the fact that causally related entities represented only 0.039% of all pairs of entities.  Applying this approach to larger report text corpora may help detect unspecified or heretofore unrecognized associations.