The Machine Learning League (MLL) objective at Moffitt Cancer Center is to advance awareness and application of Machine Learning, Deep Learning, and Artificial Intelligence (AI) across the multiple disciplines in cancer research. This will be achieved by sharing the latest advancements in the field and brainstorming on how they could be applied to solving the cancer problem. In addition, there will be educational sessions on new tools and technologies that are used in machine learning applications.
League events will include guest lectures, presentations, and tutorials, as well as, software tools, packages, and libraries.
The Machine Learning League meetings are held internally typically every other Thursday at noon EST. This is an open invitation to anyone who is interested or wishes to participate in learning and understanding about topics related to machine learning.
|May 13, 2021||Eduardo Carranza and Naveena Gorre||Latest advancements in AI discussed at NVIDIA-GTC||In our meeting, we will be discussing topics from this year’s NVIDIA GPU Technology Conference (GTC). The NVIDIA GTC conference brings together researchers, technologists, and innovators from all over the world to share their AI implementations and learn about NVIDIA’s hardware advances.||View video presentation|
|May 27, 2021||Clark MacDonald (CoreScientific)||PLEXUS Software Stack||We are excited to host Clark MacDonald from Core Scientific. He will be demoing their PLEXUS software stack. PLEXUS provides a software stack that streamlines workflows for Data Scientist and AI /ML operations.||View video presentation|
|June 10, 2021||Phil Szepietowski||Transformers and the Hugging Face Library: Tools for Modern NLP Pipelines||We are excited to be hosting a presentation by Phillip Szepietowski. He will be discussing some basic background on the transformer architecture used in many modern NLP algorithms (BERT, for example). To illustrate their use he will give an overview of the Hugging Face library which has implemented many of these algorithms in a streamlined API. Finally, he will illustrate how some of these tools are being used in current Moffitt use-cases.||
View video presentation
|June 24, 2021||Eduardo Carranza and Naveena Gorre||NVIDIA and BMW's Collaboration for the Factory of the Future||We will present BMW’s implantation of Nvidia Ominverse and Isaac Sim to bring you the factory of the future.||
View video presentation
|July 8, 2021||Larry Hall, Nathan Beach and Tiffany Ferrer||Generalization Challenges for Deep Learning for Medical Imaging: A Case Study||Please join us as we welcome Kaoutar Ben Ahmed and Lawrence O. Hall from the Department of Computer Science and Engineering at the University of South Florida who will discuss Generalization Challenges for Deep Learning for Medical Imaging: A Case Study. Is it possible to build highly accurate deep neural network models to detect COVID-19 from chest X-ray images? Tune in and find out.||View video presentation|
|July 22, 2021||Steve Gately||NVIDIA Clara - Software Tools and Frameworks for Data Science, Clara Imaging||Clara Imaging is an application framework that accelerates the development and deployment of AI in medical imaging.||View video presentation|
|August 5, 2021||Kedar Kulkarni (Head of Health Informatics)||MCAP and Advanced Analytics||Through this presentation, we will learn about the Moffitt Cancer Analytics Platform (MCAP). Specifically the reasons for moving to a cloud-based analytics platform. The journey to the cloud so far, where are we and where are we going. And finally, about the Art of the Possible with MCAP.||No recording available|
|August 19, 2021||Steve Gately||NVIDIA Clara - Software Tools and Frameworks for Data Science, Clara NLP||NVIDIA Clara NLP is a collection of models and resources that can support natural language processing and understanding workflows in healthcare and life sciences.||View video presentation|
|August 26, 2021||Jakka Ramesh Sairamesh, MPhil, PhD - CEO and President, CapsicoHealth, Inc||Driving Value-Based Care and Population Health Through AI Platforms and Real-Time Decision Tools||In the era of value-based care and rapidly transforming virtual care, our vision is to bring proven AI-driven real-time analytics for providers and payers to simplify care models and reduce costs. Our world-class AI platform enables rapid intelligence tools and FHIR interoperability on very large data sets (including claims, EMR, environmental and socio-economic factors) to enable value-based care models such as BPCI-A, CJR, OCM and Alternative or custom payment models. Our AI platform identifies opportunities and boosts productivity for financial analysts and clinicians by analyzing billions of data points and accurately identifying cohorts at-risk. CapiscoHealth’s platform drives 99% accuracy in data quality, 15% reduction in operational costs, over 80% accuracy in risk prediction (e.g. hospitalizations, emergency visits and infections), and enables teams to deploy, manage and version pathways and risk models. CapsicoHealth has launched the First Population Health and Value-Based Care Solution on Cloud Based FHIR Interoperability Platforms (such as Google Health Data Engine and HAPI based Solutions), and deployed AI Workbenches to identify high-risk populations, streamline care and help clinicians with decision-making.||View video presentation|
|September 2, 2021||Charles Donly||Making Machine Learning Radically Accessible||Since 2011, we have had a boom in compute power for machine learning: era of discovery. This era has created amazing models that can program themselves and find new discoveries. This era also has two roadblocks. First, programming (Machine learning) costs are skyrocketing (>$Millions) due to data preparation and compute time. Second, it is very difficult to explain the model that is created and thus difficult to implement in the business. Neurologix is developing a software/hardware platform that is simpler and less expensive to build ML (focused UI/UX) and virtualized GPUs.||No recording available|
|September 16, 2021||Steve Gately||NVIDIA Clara - Software Tools and Frameworks for Data Science, Genomics||NVIDIA Clara Parabricks is a computational framework supporting genomics applications from DNA to RNA. It employs NVIDIA’s CUDA, HPC, AI, and data analytics stacks to build GPU accelerated libraries, pipelines, and reference application workflows for primary, secondary, and tertiary analysis.||View video presentation|
|September 23, 2021||Jakka Ramesh Sairamesh, MPhil, PhD||Driving Value-Based Care in Oncology Through AI Models and Workbenches for Optimizing Treatment and Care Pathways, and Forecasting Costs||In the era of value-based care, our vision is to bring proven AI driven real-time analytics for providers and payors to simplify care models and reduce costs. Our operational AI workbench and models drive intelligence on very large data sets (including claims, EMR, environmental and socio-economic factors) to enable value-based care models in Oncology including Alternative payment models.||No recording available|
|September 30, 2021||Andrew Borkowski||Fastai versus Keras for Metastatic Adenocarcinoma Classification, Which one is Better?||During the presentation, I will go over the high-level APIs for PyTorch and TensorFlow using metastatic adenocarcinoma classification as an example. In addition, I will briefly describe non-coding platforms Apple CreateML and Lobe that one may use for image classification.||No recording available|
|Oct. 7, 2021||Susana Lai Yuen||Multi-objective Neural Architecture Search for Medical Image Segmentation||I will introduce our work on EMONAS-Net, an efficient multi-objective NAS framework for 3D medical image segmentation that optimizes both the segmentation accuracy and size of the network. I will present the architecture of our EMONAS-Net and a surrogate-assisted multi-objective evolutionary-based algorithm (SaMEA) that efficiently searches for the best architectures and hyperparameters. Finally, I will present the results of our model on segmenting various anatomical structures from publicly available medical image datasets. In all of the benchmarks, EMONAS-Net finds architectures that perform better or comparable with state-of-the-art NAS methods while being considerably smaller and reducing the architecture search time by more than 50%.||View video presentation|
|October 21, 2021||Steve Gately||RAPIDS – Harness the power of GPU’s to Easily Accelerate your Data Science, Machine Learning, and AI Workflows||Today, data science and machine learning has become the world's largest computer segment. Modest improvements in the accuracy of analytics models translate into billions to the bottom line. To build the best models, data scientists toil to train, evaluate, iterate, and retrain for highly accurate results and performant models. With RAPIDS™, processes that took days, now it take minutes, making it easier and faster to build and deploy value generating models.||View video presentation|
|November 11, 2021||Ian Perera||Machine Learning for Analysis of Behavior-Based and Physiological Measures at IHMC||The Florida Institute for Human and Machine Cognition presents recent work on machine learning and natural language processing methods for physiological analysis, including analysis of speech and breath for detecting hypercapnia and stress, resting-state EEG analysis for prediction of performance in fine-grained motor tasks, and natural language processing for determining stress, memory abilities, and other cognitive skills. We also present work on a novel behavioral assessment and associated data collection that tests and stresses multiple communication and cognitive skills simultaneously while providing data for the described analysis measures.||View video presentation|
|December 2, 2021||Michal Tomaszewski||Novel Radiomic Approaches for Maximized Clinical Impact||Dr. Tomaszewski received his PhD from the University of Cambridge, where he developed a new imaging technique for optoacoustic measurement of tumor vascularization. Following graduation, he joined the laboratory of Dr. Robert Gillies at Moffitt Cancer Center, where he focused on novel methods for radiological image quantification, working on several radiomic projects in MRI, MR Linac and CT. In the summer of 2021 Dr. Tomaszewski joined Merck & Co, where he works on MRI biomarker development and quantitative image analytics. In his talk, Dr. Tomaszewski will discuss the application of radiomics for clinical trial enrollment, and the rapid development of the field in MR guided radiotherapy applications.||View video presentation|
|January 6, 2022||Dr. Thomas Fuch||Clinical-grade Computational Pathology: Hype and Hope for Cancer Care||Pathology is in the midst of a revolution from a qualitative to a quantitative discipline. This transformation is fundamentally driven by machine learning in general and computer vision and deep learning in particular. This talk will focus on our recent advances in deep learning for tumor detection and segmentation, on how we train these high-capacity models with annotations collected from pathologists and how the resulting systems are implemented in the clinic.||View video presentation|
|January 20, 2022||Matthew Schabath||NSCLC Response to Therapy Utilizing AI of CT Image Analysis||Checkpoint blockade immunotherapy demonstrates durable and long-term survival benefit in 20-50% patients with advanced stage non-small-cell lung cancer (NSCLC). Patient-level response to immunotherapy response is complex and includes various phenotypes including rapid disease progression, hyperprogression (HPD), and acquired resistance. Because of the complexity and heterogeneity of response to immunotherapy, there is an urgency to identify highly predictive biomarkers that can predict treatment response and potentially stratify patients into distinct risk groups of survival and progression. This presentation will demonstrate the utility of AI in CT image analysis.||View video presentation|
|February 17, 2022||Sudeep Sarkar||Measuring Economy from Space||An interdisciplinary team from the University of South Florida (USF), the University of California, Berkeley, and Maxar developed a new artificial intelligence (AI) imagery analysis tool to derive insight about indicators of human activity to characterize the economic impact of lockdowns and reopening during the COVID-19 pandemic. Such indicators can provide insight into understanding patterns of activities such as shopping center parking lots, airports, medical facilities, schools, recreational places, and religious sites. In this talk, I will demonstrate this idea using two objects of interest – cars and airplanes as indicators of economic activity. I will share the underlying AI algorithms and show results on actual data. Our airplane detection solution won the Rapid Action Coronavirus Earth observation (RACE) upscaling challenge, sponsored by the European Space Agency and the European Commission, and now is integrated into the RACE dashboard, providing real-time information.||No recording available|
|March 17, 2022||Reza Fazel-Rezai, Mathworks||Machine and Deep Learning for Medical Imaging with MATLAB||AI techniques are increasingly seen as powerful tools to address many complex problems. In this technical talk, we'll explore in detail the workflow involved in developing and adapting machine and deep learning algorithms for medical image classification or segmentation problems using real-world case studies. Some of the tasks we'll explore in this workflow are: * Import and manage large sets of images without loading them into memory * Build networks from scratch with a drag-and-drop interface of Deep Network Designer * Perform classification tasks on images, and pixel-level semantic segmentation on images * Semi-automate ground-truth labeling efforts to increase training dataset * Understand hyperparameter tuning and why it's important||No recording available. See slides from the presentation|
|March 29, 2022||Katherine Drabiak||Legal and Ethical Issues Associated with Machine Learning and Artificial Intelligence in Healthcare*
Note: This is a Research Seminar Series so it has a different zoom link included in the details.
|Machine learning and artificial intelligence (ML/AI) promises to use algorithms to collect, curate, and process data that reveal complex interactions and patterns to improve healthcare. ML/AI offer the potential of streamlining mechanisms to screen for disease, predict differential drug responses, and provide decision-support. Considering, or implementing, ML/AI into clinical practice raises numerous legal and ethical questions, including: What are the implications for malpractice liability? How are ML/AI devices regulated, and how can we be certain that they are effective? What are potential benefits, or alternatively potential for error? What is, or should be, the interaction between AI and the clinician? How will this change the clinician-patient relationship? Please join us for this special presentation by Katherine Drabiak, JD, at the following link: https://moffitt.zoom.us/j/96736084348||View video presentation|
|April 28, 2022||Dr. Varun Ganapathi||How Machine Learning and Human-in-the-Loop Approaches are Expanding the Capabilities of Automation||Why automation efforts fall short in this complex and constantly evolving environment, how exceptions and outliers can actually make automation stronger, and how emerging machine-learning-based technology platforms, combined with human-in-the-loop approaches, are already expanding what is possible to automate across a number of industries.||View video presentation|
|May 26, 2022||Dr. Bryan Barker and Dr. Erik Chang||Oracle Open Data Platform for Researchers||Oracle Open Data (OOD) will be a platform for researchers, educators, students and developers who create, use and manipulate large data sets in their daily work. OOD will enable data producers to publish and share data easily by providing access to curated data sets. Unlike AWS, Azure and Google, which primarily offer free access to data and opportunities for data producers to monetize data, the OOD focuses on and improves the data user experience with purpose-built tools that enable users to concentrate on the results of using large data sets while controlling the extent to which they engage in data curation, data management, and compliance and security issues.||View video presentation|
|July 21, 2022||Naveena Gorre||Moffitt AI Interface for Oncology||Developing Machine Learning models for clinical applications from scratch can be a cumbersome task with varying expertise as well as seasoned developers and researchers may often face incompatible frameworks and data preparation issues. This is further complicated in the context of diagnostic radiology and oncology applications, given the heterogenous nature of the input data and the specialized task requirements. Our goal is to provide clinicians, researchers, and early AI developers with a modular and user-friendly software tool that can effectively meet their needs to explore, train, and test AI algorithms by allowing users to interpret their model results. To demonstrate this approach, we developed the Moffitt AI Interface for Oncology (MAIIO). MAIIO is based on Django web framework in Python. MAIIO/Django/Python in combination with another data manager tool/platform, XNAT/Gen3 can provide a comprehensive while easy to use machine learning tool.||View video presentation|
|Aug. 4, 2022||Dr. Felix Dietlein||Challenges and approaches for expanding cancer genomics from protein-coding to noncoding regions||Although many mutations have been characterized in 2% of the genome that encodes for proteins, the role of mutations in the noncoding genome are less well understood. I will present on our latest work to characterize mutation patterns in noncoding regions and outline perspectives for computational approaches to expand our understanding of somatic mutations in tumor development.||View video presentation|
|Sept. 15, 2022||Dr. Raymond Mak, H.M.D.||Augmenting Clinical Performance in Radiation Oncology Through AI-Human Collaboration||At the Artificial Intelligence in Medicine Program and Department of Radiation Oncology at Brigham and Women’s Hospital/Dana-Farber Cancer Institute, we have focused on developing AI algorithms for risk prediction and task automation in cancer care and developing a pipeline for clinical validation, deployment and prospective clinical trial testing. In this talk, we will review our translational approach with a focus on the 1) use of high quality, curated training data for high performance AI auto-segmentation algorithms, 2) the use of failure mode analyses with expert clinicians and 3) functional validation as opportunities to improve AI performance and human trust. We will describe our approach of testing how clinicians interact with AI segmentations and demonstrating performance improvement as a critical assessment for clinical deployment. We will also review approaches for implementing effective AI-Human work flows to complement human strengths and weaknesses (and vice versa).||View video presentation|
|Nov. 3, 2022||Dr. Berkman Sahiner
Note: This is a Research Seminar Series, so it has a different zoom link included in the details.
|Considerations for the evaluation of AI/ML-enabled medical imaging devices||Machine learning continues to be an important focus of research in medical imaging. Novel results and their practical implementations create opportunities for new types of artificial intelligence/machine learning (AI/ML)-based devices and invite new questions. AI/ML has the potential to benefit patients, caregivers and device developers, and the benefits will grow as these new questions are more comprehensively resolved. I will review the basics of regulatory pathways for devices to reach the U.S. market and talk about different types of AI/ML-enabled imaging devices available in the U.S., including devices that perform computer-aided detection, diagnosis, triage, radiological acquisition and optimization guidance, and image analysis and processing. I will also briefly discuss a paper by the FDA on a proposed regulatory framework and request feedback for modifications to AI/ML-based software as a medical device. Additionally, I present some of the gaps in AI/ML device design and evaluation from a regulator's perspective. Finally, I will describe some of the research projects in our office to tackle these gaps and conclude my remarks on the total product life cycle for AI/ML-enabled devices.
Please join us for this special presentation by Dr. Berkman Sahiner, PhD at https://moffitt.zoom.us/j/94147091733
|View video presentation|
|Nov. 10, 2022||Jakka Ramesh Sairamesh, MPhil, PhD - CEO and President, CapsicoHealth, Inc.||Awaiting||Awaiting||Awaiting presentation|
|Jan. 5, 2023||Dr. David Joon Ho, Memorial Sloan Kettering Cancer Center||Deep learning-based whole slide image segmentation for efficient, objective, and reproducible assistance in pathology||Pathology plays a crucial role to diagnose cancer and to assess its progression from hematoxylin and eosin (H&E)-stained tissue samples. Diagnosis and assessment of tissue samples on glass slides have been done under microscopes which can be inefficient and subjective. Digitization of glass slides and deep learning-based computational approaches have been investigated to help this process. Especially, semantic segmentation, also known as pixel-wise classification, of whole slide images quantifying multiple tissue subtypes is a prerequisite step for clinical interpretations. In this talk, Dr. Ho will present how tissue segmentation can be done using deep learning and assist pathologists in an efficient, objective, and reproducible way. Specifically, (1) Dr. Ho will describe Deep Multi-Magnification Network for multi-class tissue segmentation which looks at morphological features from multiple magnifications for more accurate segmentation. (2) He will explain Deep Interactive Learning to efficiently annotate whole slide images to train segmentation models. (3) He will introduce two clinical applications where pathologists can be assisted by tissue segmentation. A breast model segmenting cancer can efficiently screen potential malignant margin slides with high sensitivity and highlight cancer regions where margins for lumpectomy are mostly benign. An osteosarcoma model segmenting viable tumor and necrotic tumor can reproducibly estimate case-level necrosis ratio from multiple slides for pre-operative treatment response assessment and can predict patient survival outcomes. In conclusion, tissue segmentation models can provide efficient, objective, and reproducible support to pathologists.||Awaiting presentation|
|Jan. 19, 2023||Dr. Keyvan Farhini||The Imaging Data Commons and Data Science at NCI||In recent years, the National Cancer Institute has made significant investments in developing a data science infrastructure in support of cancer research. This presentation highlights developments in NCI’s Cancer Research Data Commons (CRDC) , with emphasis on the Imaging Data Commons (IDC), and describes how the research community may engage to make use of, and contribute to, these resources.||Awaiting presentation|
|Feb. 2, 2023||Dr. Michael J Donovan, MD, PhD PreciseDx, CMO and Professor of Pathology Mt. Sinay Medical Center||Awaiting||Awaiting||Awaiting presentation|
|Feb. 16, 2023||Charles Donly||Awaiting||Awaiting||No recording available|
|March 2, 2023||Not Confirmed||Not Confirmed||Not Confirmed||Not Confirmed|
|March 16, 2023||Dr. Andrew D. Smith, MD, PhD||Augmented Intelligence in Evaluating Advanced Cancer Response to Therapy||This presentation reviews the development and validation of an FDA-cleared augmented intelligence CT and MRI viewer and reporting system for evaluation of advanced cancer response to therapy. Learn how the combination of 1) guided workflows that adhere to best practice standards, 2) artificial intelligence (AI) algorithms for tumor measurement, anatomic labeling, and longitudinal tracking, and 3) automated reporting can reduce errors, increase efficiency and standardization, improve communication, and facilitate widespread radiomics analysis for advanced cancer.||Awaiting Presentation|
|March 30, 2023||Tameem Samawi||kBonsai: Your One Stop Data Shop||kBonsai is an invite-only data platform for researchers working in Machine Learning Medical Research. The platform provides its users diverse, high-quality datasets to be used to validate or train predictive Machine Learning Models. Datasets are curated by data scientists and healthcare professionals from the highest quality providers in the world, and we work directly with clients to ensure their data needs are met. For specific requests, users are able to post RFPs through the platform to a network of data providers who typically respond in 24-48 hours. Our goal is to make data curation the least time consuming aspect of Machine Learning projects in an effort to shorten the R&D lifecycle and enable teams to focus on research itself.||Awaiting Presentation|
|April 13, 2023||Dr. Reza Fazel-Rezai and Renee Qian||Medical Image Analysis and AI Workflows in MATLAB||Medical images come from multiple sources such as MRI, CT, X-ray, ultrasound, and PET/SPECT. The challenge is to visualize and analyze this multi-domain image data to extract clinically meaningful information and conduct other tasks such as training AI models. MATLAB provides tools and algorithms for end-to-end medical image analysis and AI workflows – I/O, 3D visualization, segmentation, labeling and analysis of medical image data. This webinar shows the complete medical image analysis workflow for AI applications. You will learn how to import visualize, segment and label medical image data and utilize these data in AI model training.||Awaiting Presentation|
If you are interested in joining the Machine Learning League distribution list or have any other questions, concerns, or even suggestions please contact Machine Learning at MachineLearning@moffitt.org.