Despite the growing popularity of deep learning neural networks for various medical imaging applications, the vast majority of algorithms to date represent early proof-of-concept designs that will require a degree of evolution before achieving practical clinical utility. In this talk, we explore several different ways in which the current generation of deep learning applications can be advanced including: 

(1) Reformulating the question: in medicine, there are often times more than one way to ask the same question---how do we reformulate a task in a way that  both maximizes clinical utility and also best leverages the strength of various deep learning algorithms?

(2) Customizing deep learning algorithms: what are the unique technical challenges posed by medical imaging data and how we design custom deep learning architectures to account for them?

(3) Clinical implementation: what are some practical experiences learned from implementing deep learning tools in the clinical setting, and what are some regulatory hurdles that will need to be considered?

In this talk, I will present the euCanSHare European project, which will develop the first centralised, secure and sustainable platform for enhanced cross-border data sharing and multi-cohort personalised medicine research in cardiology. This web-platform will be made open-access for researchers worldwide to promote novel big data and multi-omics approaches. At its heart, the platform will contain the most comprehensive cardiovascular data catalogue ever assembled, which will facilitate data discoverability and exploitation in full alignment with the so-called FAIR principles. Initially populated with 35 European and Canadian cohorts (corresponding to about one million records), the platform will provide extensive functionalities, including for data deposition, data harmonization and quality control, which will support the integration of new cohorts beyond the duration of the project, thus ensuring its scalability. The features of the platform will be demonstrated and adjusted through several use cases, including for biomarker validation, knowledge discovery, cardiovascular risk assessment, public health research and industry-driven studies. euCanSHare’s legal framework will be established through detailed ethical and legal interoperability analysis, while investigating innovative solutions for promoting responsible Open Science based on the emerging blockchain technology.

While disease detection and diagnosis utilize a variety of tests, interventional procedures in these patients (including therapeutic assessment and monitoring) have largely used imaging alone. Developing computerized schemes to assist the clinician in this regard requires “unlocking” all embedded information captured by different Big Data streams. Creating accurate predictors for this purpose also necessitates evaluation of advanced imaging features (termed “radiomic” features) together with histology or molecular data to improve tissue characterization and biological quantitation in vivo.

The research in my group focuses on developing novel radiomics and machine learning tools for disease diagnosis, intervention, and treatment evaluation. This involves integrating information across multiple length scales of Big Data by spatially resolving imaging (macro-scale) with molecular and pathology (micro-, nano- scales) data. Specific problems include: (a) decision support for treatment (e.g. choice of therapy), (b) targeting therapeutic procedures (e.g. guiding ablation, radiotherapy, surgery), and (c) biological quantitation for disease characterization in vivo. We have also evaluated the impact of different sites, scanners, and acquisition parameters when developing image analytics tools and radiomic features. These will be discussed in the context of clinical applications in colorectal, renal, and prostate cancers, as well as digestive diseases.

In 2017, the Department of Radiology initiated a new diversity program, which aims to support the inclusion of qualified women and underrepresented minorities in our Department and in our leadership teams, specifically. Considering the complexity of diversity in terms of race/ethnicity, economic background, age, gender, sex and sexuality, among others, we seek to create a fundamental framework that allows us to ensure equal opportunities for every member of our community. Introducing powerful advocates at the leadership table reduces the risk of bias and discrimination at the workplace and provides our trainees and faculty with a diverse set of role models.

This presentation will discuss the current status of diversity in Radiology, ongoing activities of the Radiology diversity program, and strategic plans for the future. The Radiology diversity program is composed of a faculty diversity committee with representatives from each section within the Department as well as a trainee diversity committee with representatives for Radiology residents, clinical fellows and research fellows. Our ongoing initiatives provide many opportunities for participation in discussion groups, social gatherings, research activities, community projects and outreach programs. Our diversity initiative ultimately seeks to unite diverse minds to inspire medical innovation: Our experiences, abilities and perspectives are unique. Nobody can use them as we can. We want to inspire each other to unleash our collective creativity and create a culture of growth, fulfillment and prosperity that far exceeds what every one of us could have achieved alone. Diversity is critical for our ability to serve patients in a multi-cultural environment, to provide inspiring role models for our trainees, to unfold discoveries at the interface of different disciplines, to address the many challenges in our health care system and to cure humanity – one patient at a time.

The University of Pittsburgh Department of Biomedical Informatics (DBMI) seeks to unlock the potential of machine learning and artificial intelligence for healthcare by building pipelines of high impact data from pathology imaging. This seminar will describe the experience of the University of Pittsburgh School of Medicine and UPMC (our health system partner) in creating large scale data to enable the use of AI in personalized medicine and immunotherapy with a focus on immuno-oncology. The lessons learned from DBMI’s experience with the NCI’s Information Technology for Cancer Research (ICTR) program with the Text Information Extraction System (TIES) Cancer Research Network (TCRN) will be featured.

The objectives are:
1) To understand the Text Information System (TIES) Cancer Research Network (TCRN)
2) To understand Whole Slide Imaging (WSI) and Computational Pathology (Comp Path)
3) To understand the role of clinical informatics and the cancer registry in enabling Pathology Imaging Infrastructure Enabled by TCRN
4) To understand Computational Pathology's role in Predictive Analytics for Precision Oncology and how this leads to better health and discovery science

The greatest issue in global breast screening is not accuracy, but a workforce crisis. Traditional computer-aided detection (CAD) for mammography decision-support could not reach the level of an independent reader. Deep learning (DL) outperforms CAD and is close to surpassing human performance. DL is already capable of decision support and density assessment for 2D full-field digital mammography (FFDM), and is on the cusp of providing consistent, accurate and interpretable mammography reading as an independent reader.

Tumor heterogeneity plays a critical role in tumor evolution.  Our prior in vitro studies highlighted the extent to which microenvironmental forces, such as hypoxia, influence this evolution.   Up until recently, a lack of approaches for analyzing the spatial patterning of molecular phenotypes has made it challenging to examine the influence of tumor micro-heterogeneity on clinical outcomes. We recently developed an approach to study multiplexed immunohistochemistry (IHC) data to establish NSCLC histology-specific molecular subtypes and identify regions of hypoxia in tumor tissues. We were then able to examine the impact of hypoxia on protein networks in situ.  In addition, we were able to use our new computational approaches to stratify patient outcomes, noting that spatially derived features better predicted 5-year survival (AUC = 0.72) compared to non-spatially derived features (AUC = 0.49) in a held-out test set. Additionally, the total area of defined hypoxic regions was better able to separate the patient population in a Kaplan-Meier survival analysis (p=0.024) compared to standard percent positivity measures (p=0.268). Our results highlight the utility of highly multiplexed spatial analysis in understanding complex cancer phenotypes and demonstrate the clinical significance of the spatial organization of phenotypic cancer drivers.

Precision medicine is the future of health care: please watch the animation at https://vimeo.com/241154708 . As a technology-intensive and -dependent medical discipline, oncology will be at the vanguard of this impending change. However, to bring about precision medicine, a fundamental conundrum must be solved: Human cognitive capacity, typically constrained to five variables for decision making in the context of the increasing number of available biomarkers and therapeutic options, is a limiting factor to the realization of precision medicine. Given this level of complexity and the restriction of human decision making, current methods are untenable. A solution to this challenge is multifactorial decision support systems (DSSs), continuously learning artificial intelligence platforms that integrate all available data-clinical, imaging, biologic, genetic, cost-to produce validated predictive models. DSSs compare the personalized probable outcomes-toxicity, tumor control, quality of life, cost effectiveness-of various care pathway decisions to ensure optimal efficacy and economy. DSSs can be integrated into the workflows both strategically (at the multidisciplinary tumor board level to support treatment choice, eg, surgery or radiotherapy) and tactically (at the specialist level to support treatment technique, eg, prostate spacer or not). In some countries, the reimbursement of certain treatments, such as proton therapy, is already conditional on the basis that a DSS is used. DSSs have many stakeholders-clinicians, medical directors, medical insurers, patient advocacy groups-and are a natural consequence of big data in health care. Here, we provide an overview of DSSs, their challenges, opportunities, and capacity to improve clinical decision making, with an emphasis on oncology and radiomics.

This presentation will be an overview of the Research Technology vision and currently available infrastructure and services. I will also discuss the unification of the IT teams from the Stanford School of Medicine and Stanford Health Care and its implications for the research community.

Prostate Cancer exhibits different clinical behavior, ranging from indolent to lethal disease. A critical clinical need is identifying characteristics that distinguish indolent from advanced disease to direct treatment to the latter. The recent renaissance of artificial intelligence (AI) research uncovered the potential of AI to improve clinical decision making. In this seminar, we will go through the potential of AI to enhance the diagnosis and the prognosis of prostate cancer using magnetic resonance images, clinical data, and histology images. We will stress the challenges and benefits of having such AI-based solutions in clinical routine. 

This presentation will describe some of the most important considerations involved in creating algorithms in medical imaging from inception to deployment as well as continued model improvement and/or monitoring. Examples of experience to date from the OSU laboratory for augmented intelligence in imaging will be provided. New paradigms in model creation and the role of image challenge competitions will also be covered. Current issues with model validation and generalizability will also be introduced as well as considerations for future work in this area.