Ipek Oguz, Ph.D. - US grants

PROJECT SUMMARY/ABSTRACT Huntington's disease (HD) is a neurodegenerative disease where brain abnormalities can be detected via MRI studies one to two decades prior to clinical diagnosis. Sensitive outcome measures are needed for enabling clinical trials during pre-manifest HD with the goal of intervention and treatment at the earliest stage possible. The PREDICT-HD study (NS040068) identified longitudinal alterations in the basal ganglia and cortical gray matter atrophy as the primary neuroimaging findings in pre-manifest HD patients. However, measurement noise is a serious concern as it can affect the ability to detect abnormalities early in the disease progression. The objective of this proposal is to develop innovative methods for quantifying the cerebral cortex and the basal ganglia in a temporally and spatially consistent manner and to leverage these techniques to improve the quantification of HD progression in patients from the existing PREDICT-HD database. We hypothesize that more accurate quantification will provide more sensitive measures of HD progression, leading to increased sensitivity to longitudinal changes prior to clinical diagnosis. The quantification is expected to be substantially more accurate than currently possible due to our novel temporal- and spatial- context-aware segmentation strategy, which leverages the inherent redundancy of longitudinal MRI data. Three specific aims will be fulfilled: Aim 1. Develop and validate a novel temporally and spatially consistent segmentation method for quantification of the basal ganglia in longitudinal studies of HD. The impact is expected to be especially large for structures with weak boundaries, such as the nucleus accumbens, which are hard to quantify with existing approaches. Validation will be accomplished via comparison with expert manual segmentations. Aim 2. Develop and validate a novel longitudinal cortical surface reconstruction method for temporally consistent cortical thickness quantification in longitudinal studies of HD. Our approach will utilize temporal image-to-image context while avoiding over-regularization. The validation will be based on reproducibility in test-retest scans and statistical discrimination power in population studies, using public datasets. Aim 3. Assess the increase in statistical sensitivity of imaging measures derived from our new segmentation approaches in a longitudinal pre-manifest HD cohort, and validate these imaging measures by documenting their association with known clinical outcome assessments (COA's) and genetic variables. We will use 1246 scans from Predict-HD to evaluate the sensitivity of developed methods and validate against clinical variables. We anticipate the proposed segmentation methods to substantially increase the sensitivity of existing imaging- based measures in HD. This will provide a means of developing and evaluating early therapeutic intervention strategies in order to prevent disease onset and slow disease progression. Equally significant, the innovative methods to be developed in this proposal are expected to be crucially important for increased sensitivity for other neurodegenerative disorders such as Alzheimer's disease or Parkinson's disease.

? DESCRIPTION (provided by applicant): The International Symposium on Biomedical Imaging (ISBI) is a forum for researchers interested in the computational and modeling aspects of biomedical imaging. The focus emphasizes methodologies that have the potential to be applicable to multiple imaging modalities and to imaging at different scales. Topics include physical, biological and statistical modeling of biological and anatomical structures, image formation and reconstruction, computational image analysis, statistical image analysis, visualization, and image quality assessment. The meeting aims to facilitate cross-fertilization of methodologies among different imaging modalities and scales, with applications ranging from the nano, molecular and cellular levels through small- animal imaging to macroscopic and whole body clinical systems. Whereas many medical imaging meetings focus on particular modalities, ISBI includes a diversity of methodologies applied to biological and anatomical imaging modalities, either emerging or well-established. Imaging applications of interest include neuroimaging, retinal imaging, functional brain mapping, effects of aging, cell biology, organization and function, gene expression mapping, drug discovery and delivery, molecular imaging, cardiovascular imaging, and cancer imaging. A principal goal of the ISBI conference is to connect image acquisition and processing methodologies with important medical and biomedical applications, from microscopic to macroscopic scales. The ISBI 2016 edition, to be held from April 13th to 16th, 2016 in Prague, Czech Republic, will be the 13th in a series of meetings co-sponsored by two IEEE societies: the Signal Processing Society (SPS) and the Engineering in Medicine and Biology Society (EMBS). Historically, ISBI conferences have had a major impact on helping students and junior faculty mature into leaders in the medical imaging field. The success of ISBI and its impact on the field, and indeed the future of the field itself, relies on training young investigators to work with state-of-the-art computational or modeling tools as well as on informing the research community of the latest progress in biomedical imaging protocols and modalities. In addition, the inherently interdisciplinary nature of the biomedical imaging field means that no single professional organization has the majority of potential participants as its members. In this context, the tutorial series, covering diverse topic and offered on the first day of the conference, is an important component, which has been very popular in previous ISBI meetings. This year, for the first time, a Women in Medical Imaging mentoring/networking event will be hosted at ISBI to support women scientists, who are underrepresented in the medical imaging field. This proposal requests funds to provide travel/registration support for US-based young investigators (graduate students and postdoctoral fellows) to present their work at ISBI 2016, reach an international audience, and have a chance to develop international linkages and collaborations.

The NSF Convergence Accelerator supports use-inspired, team-based, multidisciplinary efforts that address challenges of national importance and will produce deliverables of value to society in the near future.

This project, Convergence Accelerator Track D: Scalable, TRaceable Ai for Imaging Translation: Innovation to Implementation for accelerated Impact (STRAIT I3), addresses fundamental gaps between the science and the engineering that is preventing the effective use of AI models with medical imaging data. The project will leverage the large number of open dataset efforts available for medical imaging, including imaging resources for COVID-19. Thousands of AI models are also published in the scientific literature each year for such data. Yet, these resources are not consistently accessible at scale nor are they able to be validated for clinical application. This project includes three thrust areas to address this problem. Thrust Area 1 democratizes access to data sets through traceable data annotation. Thrust Area 2 transforms the assessment and peer review process for data, to ensure fair and consistent evaluation of technologies. Thrust Area 3 targets reproducible execution and comparison of models to facilitate translation to practice. In Phase I, this Convergence Accelerator project will create direct public health and technology benefits by enhancing the radiological assessment of COVID-19 pneumonia. In Phase II, it will extend these benefits into a medical imaging ecosystem spanning multiple medical imaging domains. Broader impacts will be achieved by engaging various identified communities through professionally led studios, consented A/B testing studies, and structured outreach. All project thrusts utilized open software and commodity hardware, wherever possible, so that the innovations from this project on scalable image data validation will enhance other related efforts in open source software, open science, reproducible science, and findable science.

This project works towards achieving a fundamental rethinking in how model-centric AI could be validated and translated in medical imaging, algorithm design, and medical science. The intellectual activities are organized around three research thrusts, each addressing an essential challenge that currently confronts the development and translation of AI-based medical imaging tools. One research thrust is on creating a lightweight data provenance and annotation interface compatible with both clinical imaging and research studies. The second is on facilitating rapid innovation in AI architectures while creating an enhanced validation/peer review process to avoid irreproducible implementations and overtraining of models. The third thrust is the integration of these efforts into a novel Model Zoo to provide robust capabilities for validation, assessment, and translation. This research effort will utilize core scientific innovations from the collaborative team consisting of members from a university (Vanderbilt), a medical center (Vanderbilt Medical Center), two industry partners (MD.ai, Kaggle), and a professional society (SIIM), alongside widely used, open source platforms. In Phase 1 of the Convergence Accelerator, the project will focus on newly created public and private datasets for COVID-19. Phase II will scale this approach to different medical imaging modalities, including dermatology and ophthalmology.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Medical images, such as computed tomography (CT) and magnetic resonance imaging (MRI) scans, are routinely transformed and analyzed using computers and software. Medical images are increasingly processed and analyzed with artificial-intelligence and machine-learning methods, such as deep neural networks. In such safety-critical domains, stringent guarantees on the behaviors of these machine-learning methods are essential, but many recent studies have shown risks with these methods, such as lack of robustness and bias. Neural-network formal verification is emerging as an approach to provide guarantees on these machine-learning models to precisely characterize their behaviors. This project's novelties are to develop a neural-network formal-specification and -verification framework for medical-image analysis tasks, apply it to ensure specifications across the medical-image-analysis technology stack, and evaluate it on two specific medical-imaging analysis tasks. The first image-analysis task is the segmentation of brain lesions from MRIs of multiple-sclerosis patients, which is the process of automatically characterizing different regions of the MRIs into corresponding anatomic structures. The second image-analysis task is image synthesis for denoising optical-coherence-tomography (OCT) scans of the retina. The project's impacts are to enable medical-imaging analysis by enhancing confidence in the machine-learning models through formal verification. Beyond the development and application of these formal methods to medical-imaging analysis, the results of this project may enhance trustworthiness in other domains, such as perception and sensing components in autonomous systems.<br/><br/>Most neural-network verification methods developed so far are applicable only to simple computer-vision tasks, such as image classification, whereas medical image analysis typically requires more sophisticated generative computer-vision methods to solve more sophisticated tasks, such as semantic segmentation, instance segmentation, and image synthesis. Developing formal methods for these generative computer-vision tasks in the context of medical-imaging analysis is the core of this project. The first major objective of the project is to develop a robustness-specification framework, building on robustness to adversarial perturbations, specification mining, and metrics from computer vision used for generative tasks. The second major objective is to develop the formal-verification methods, building on reachability analysis of neural networks, specifically developing reachability methods for up-sampling layers used in generative models. The third major objective is to consider the robustness of generative models for image synthesis, such as those trained through generative-adversarial-network (GAN) processes. The fourth major objective is to evaluate these specification and verification methods on the MS lesion segmentation and OCT image-synthesis denoising tasks. The researchers will organize relevant competitions and challenges, such as continuing the Verification of Neural Networks Competition (VNN-COMP) and the IEEE ISBI Longitudinal MS Lesion Segmentation Challenge, and will develop benchmarks for the formal-methods, machine-learning, computer-vision, and medical-imaging-analysis research communities based on the research and results of this project.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.