AI based methods are feeding all the innovations Philips is bringing to the market as clinical and technical solutions. I wish to share the perspective of using AI for active real time monitoring of MR systems status and for helping technologists and radiologists in their daily routines and workflows
Short Bio
I am a Biomedical Engineer. I worked in clinical and research facilities for few years after my Master Degree. Then I moved into the biomedical industry and I have been working for Philips for almost 20 years as MR clinical scientist supervising, supporting and facilitating the MR related research activity done in Italy on Philips MR platforms by our customers.
We describe an approach for incorporating prior knowledge into machine learning algorithms. We aim at applications in physics and signal processing in which we know that certain operations must be embedded into the algorithm. Any operation that allows computation of a gradient or sub-gradient towards its inputs is suited for our framework. We derive a maximal error bound for deep nets that demonstrates that inclusion of prior knowledge results in its reduction. Furthermore, we show experimentally that known operators reduce the number of free parameters. We apply this approach to various tasks ranging from computed tomography image reconstruction over vessel segmentation to the derivation of previously unknown imaging algorithms. As such, the concept is widely applicable for many researchers in physics, imaging and signal processing. We assume that our analysis will support further investigation of known operators in other fields of physics, imaging and signal processing.
Short Bio
Prof. Dr. Andreas Maier was born on 26th of November 1980 in Erlangen. He studied Computer Science, graduated in 2005, and received his PhD in 2009. From 2005 to 2009 he was working at the Pattern Recognition Lab at the Computer Science Department of the University of Erlangen-Nuremberg. His major research subject was medical signal processing in speech data. In this period, he developed the first online speech intelligibility assessment tool â PEAKS â that has been used to analyze over 4.000 patient and control subjects so far. From 2009 to 2010, he started working on flat-panel C-arm CT as post-doctoral fellow at the Radiological Sciences Laboratory in the Department of Radiology at the Stanford University. From 2011 to 2012 he joined Siemens Healthcare as innovation project manager and was responsible for reconstruction topics in the Angiography and X-ray business unit. In 2012, he returned the University of Erlangen-Nuremberg as head of the Medical Reconstruction Group at the Pattern Recognition lab. In 2015 he became professor and head of the Pattern Recognition Lab. Since 2016, he is member of the steering committee of the European Time Machine Consortium. In 2018, he was awarded an ERC Synergy Grant â4D nanoscopeâ. Current research interests focuses on medical imaging, image and audio processing, digital humanities, and interpretable machine learning and the use of known operators.
Metabolomics by NMR is ideally suited for untargeted research and unsupervised analysis. I will show by way of examples various statistical methods used to get the most from NMR data, including some developed by us, as well as an example of a machine learning approach to predict the chemical shifts of metabolites in urine samples, allowing for automated assignment of urine spectra.
Short Bio
Full Professor of Chemistry at the University of Florence, co-founder and present Director of the Center of Magnetic Resonance (CERM), co-founder and present Director and President of the Interuniversity Consortium on Magnetic Resonance of Metallo Proteins (CIRMMP). His research interests include bioinorganic chemistry, structural biology, development of NMR-based structural methodologies in solution and in the solid state, metalloproteins and metalloenzymes, spectroscopy, theory of electron and nuclear relaxation, NMR of paramagnetic species, relaxometry, contrast agents and NMR-based analytical methods in general. Particularly worth mentioning are his recent studies on the integration of structural techniques, where he has shown that simultaneous refinement of X-ray and NMR data can lead to i) a significant increase of accuracy of the resulting structural data and ii) a robust strategy to assess if structural differences exist for a biomolecule between solution and crystalline state. He has also significantly contributed to the theoretical understanding of the phenomenon of Dynamic Nuclear Polarization (DNP), both in solids and in liquids. DNP is a promising strategy to enhance the sensitivity of the NMR experiment, possibly transforming NMR into a completely new technique for the investigation of biological systems in ways that would have been unconceivable only a few years ago. Starting from 2008, his research has been also directed towards metabolomics, aiming at obtaining the metabolic profiles of biological fluids such as urines and blood using NMR spectroscopy; defining procedures for sample preparation and for the acquisition of NMR spectra; developing statistical methods for the analysis of the data; and correlating the metabolic profiles with pathophysiological characteristics of the subjects studied. The results of this research, published in international journals, have been cited several times not only in specialized literature but also in other sectors, in books, websites and also in non-specialist press. Claudio Luchinat has become one of the international reference points for metabolomics, and is invited to give plenary lectures on the topic at conferences both specialized and in other sectors. He is invited as a keynote speaker from research laboratories in various countries around the world, and invited to inaugurate new research centers and research infrastructures.
Artificial Intelligence in medical imaging involves research in task-based discovery, predictive modeling, and robust clinical translation. Quantitative radiomic analyses, an extension of computer-aided detection (CADe) and computer-aided diagnosis (CADx) methods, are yielding novel image-based tumor characteristics, i.e., signatures that may ultimately contribute to the design of patient-specific cancer diagnostics and treatments. Beyond human-engineered features, deep convolutional neural networks (CNN) are being investigated in the diagnosis of disease on radiography, ultrasound, and MRI. The method of extracting characteristic radiomic features of a lesion and/or background can be referred to as âvirtual biopsiesâ. Various AI methods are evolving as aids to radiologists as a second reader or a concurrent reader, or as a primary autonomous reader. This presentation will discuss the development, validation, and ultimate future implementation of AI in the clinical radiology workflow including the example of breast cancer.
Short Bio
Maryellen Giger, Ph.D. is the A.N. Pritzker Distinguished Service Professor of Radiology, Committee on Medical Physics, and the College at the University of Chicago. She has been working, for multiple decades, on computer-aided diagnosis /machine learning/deep learning in medical imaging and cancer diagnosis / management. Her AI research in breast cancer for risk assessment, diagnosis, prognosis, and therapeutic response has yielded various translated components, and she is using these âvirtual biopsiesâ in imaging-genomics association studies. She has now extended her AI in medical imaging research to include the analysis of COVID-19 on CT and chest radiographs, and is PI on the NIBIB-funded Medical Imaging and Data Resource Center (MIDRC). Giger is a former president of AAPM and of SPIE; and is the Editor-in-Chief of the Journal of Medical Imaging. She is a member of the National Academy of Engineering; Fellow of AAPM, AIMBE, SPIE, SBMR, IEEE, IAMBE and COS; and was cofounder, equity holder, and scientific advisor of Quantitative Insights [now Qlarity Imaging], which produces QuantX, the first FDA-cleared, machine-learning driven CADx system.
Keywords
transfer learning, radiomics, computer-aided diagnosis, machine learning, breast cancer MRI
Resting-state functional magnetic resonance imaging (rs-fMRI) has been successfully employed to understand the organisation of the human brain. For rs-fMRI analysis, the brain is typically parcellated into regions of interest (ROIs) and modelled as a graph where each ROI is a node and pairwise correlation between ROI blood-oxygen-level-dependent (BOLD) time series are edges. Recently, graph neural networks (GNNs) have seen a surge in popularity due to their successes in modelling unstructured relational data. The latest developments with GNNs, however, have not yet been fully exploited for the analysis of rs-fMRI data, particularly with regards to its spatio-temporal dynamics. Herein we present a novel deep neural network architecture, combining both GNNs and temporal convolutional networks (TCNs), which is able to learn from the spatial and temporal components of rs-fMRI data in an end-to-end fashion. In particular, this corresponds to intra-feature learning (i.e., learning temporal dynamics with TCNs) as well as inter-feature learning (i.e., leveraging spatial interactions between ROIs with GNNs). We evaluate our model with an ablation study using 35,159 samples from the UK Biobank rs-fMRI database. We also demonstrate explainability features of our architecture which map to realistic neurobiological insights. We hope this model could lay the groundwork for future deep learning architectures focused on leveraging the inherently and inextricably spatio-temporal nature of rs-fMRI data.
Short Bio
Computer Science PhD student at the University of Cambridge, under the supervision of Prof. Pietro Lio’. Interested in applications of Machine Learning in real-world datasets.
To date, many examples of machine learning models that successfully discriminate between patients with psychiatric disorders and healthy individuals based on neuroimaging data have been published. However, diagnostic accuracy and generalisability are not yet good good enough for clinical application. The limitations of the current learning designs are discussed with a focus on the weak relationship between input and output. A number of approaches that may lead to prediction models that have improved performance and clinical relevance are discussed, including prognostic (as opposed to diagnostic) and normative modeling.
Short Bio
From September 1995 I have worked as a post-doc in the Neuroimaging section of the Department of psychiatry of the University Medical Center Utrecht, and from 1998 onwards as an assistant professor. During the first years, I developed the image processing and analysis system of the group. As of 2015, thousands of scans have been analyzed using this pipeline. Gradually my research interests moved from image analysis of structural MRI brain scans to (longitudinal) brain data analysis and modeling, and from group-level statistics to individual predictions using Machine Learning. My research involves the study of both healthy development and psychiatric disorders including schizophrenia and bipolar disorder. Currently, the focus of my research line is on developing individual prediction models based on MRI brain image data and clinical data from psychiatric patients, with special attention to interpretable models. Starting in 2016, part of my research and teaching activities is at the Faculty of Humanities of Utrecht University.
In this presentation, RF pulses designed by deep reinforcement learnining will be introduced. A newly developed algorithm, DeepRF, demonstrates successful generation of various types of RF (e.g, excitation, inversion, B1-insensitive inversion âŠ) by self-training. The resulting RFs reveal improved SAR while pertaining slice profiles when compared to conventional SLR or adiabatic RF pulses.
Short Bio
Jongho Lee is Associate Professor at the Department of Electrical and Computer Engineering, Seoul National University. He received his Ph.D in Electrical Engineering and Ph.D minor in Psychology at Stanford University (2007). From 2007 to 2010, he worked at National Institutes of Health as a research fellow. From 2010 to 2014, he continued his academic career as Assistant Professor at the Department of Radiology, University Pennsylvania. In 2014, he moved back to Korea to join a faculty position at Seoul National University. His research interests include development of neuroimaging methods and imaging devices.
Keywords
Deep Reinforcement Learning, RF design, Self learning, DeepRF, Deep Learning and deep neural network
Deep learning (DL) techniques have emerged as an alternative for accelerated MRI due to their improved reconstruction quality. State-of-the-art methods use a physics-guided approach, incorporating the multi-coil encoding operator. Such physics-guided neural networks are trained end-to-end, typically in a supervised fashion using fully-sampled/high-quality ground-truth references. However, in a number of cases, it is impossible to acquire ground-truth data, hindering the applicability of the DL methods. In this talk, we present recent advances in self-supervised learning for physics-guided DL reconstruction, when ground-truth data is not available. We show that the self-supervised DL reconstruction trained on sub-sampled data performs similar to the supervised approach trained on ground-truth data. We also show applications in absence of fully-sampled data, extending the utility of physics-guided DL reconstruction.
Short Bio
Mehmet Akcakaya received the B. Eng. degree from McGill University, Montreal, QC, Canada, in 2005, and the S.M. and Ph.D. degrees from Harvard University, Cambridge, MA, USA, in 2010. He is an associate professor at the University of Minnesota, Minneapolis, MN, USA. His work on accelerated MRI has received a number of international recognitions. He holds an R01 Award and a Trailblazer Award from NIH, and a CAREER Award from NSF. His research interests include image reconstruction, machine learning, MRI physics, inverse problems and signal processing.
Keywords
Deep learning, Self-supervision, Unsupervised training
Premature birth induces modifications in the developmental trajectory of the brain during a period of intense maturation with possible lifelong consequences. We performed anatomical and functional MRI at term-corrected age on 88 newborns with varying gestational age (GA) at birth. We obtained measures of resting-state functional connectivity, functional connectivity density, local functional activity and regional tissue volume in a set of 90 cortical and subcortical brain regions. A data-driven multivariate analysis framework (i.e. Machine Learning framework) was built to exploit the high dimensionality of the data in assessing the sensitivity of each metric to the effect of premature birth. The results showed that prematurity was associated with bidirectional alterations of functional connectivity and regional volume, and, to a lesser extent, with modification of regional activity. Notably, the effects of prematurity on functional connectivity were spatially diffuse, whereas effects on regional volume and activity were more localized to specific regions, such as subcortical structures. Machie Learning methods appear well suited to identifying premature infants at risk of negative neurodevelopmental outcome based on MR neuroimaging.
Short Bio
Dr Chiarelli received his BS and MS in Physics Engineering at the Polytechnic of Milan. He received his PhD in Neuroimaging at the University of Chieti. He worked as a Post Doc at Beckman Institute, IL, USA. From 2017 he is an Assistant Professor in Applied Physics at the University of Chieti. His research focuses on optical imaging, MRI, EEG, signal processing and machine learning for studying the human brain. He also has interest in machine learning for clinical diagnosis and prognosis.
Keywords
Prematurity, Magnetic Resonance Imaging, Machine Learning, Functional MRI, Volumetric MRI
This talk describes the current and the (possible) future regulatory framework of explainability. After an overview on the GDPR âright to explanationâ, product safety law and informed consent, it then describes why ethical guidelines are not sufficient. It then moves on to a comparative discussion about new proposals on AI regulation at the European. The EU and the Council of Europe have proposed broad horizontal regulations (affecting every sector) based on basic principles like non-discrimination and transparency, explainability and accountability. A comparative analysis shows that these proposals both contain similar core elements: a duty to inform about the use of AI, about the capabilities and limits of an AI system and the duty to make AI explainable Importantly it has been recognized that there are technical limits to explainability and that explainability must be balanced against other interests. Besides explainability, documentation and audits/impact assessments will be core elements. The talk concludes with open questions and an outlook on upcoming policy developments.
Short bio
Mag. iur. David Schneeberger, BA BA MA is a research and teaching assistant at the chair of medical law at the University of Vienna and a project team member at the Medical University of Graz (FWF project: Reference Model of Explainable AI for the Medical Domain). He is currently working on a PhD thesis on “The use of Machine Learning in public administration and the role of the duty to state reasons: An attempt to defuse this dichotomy.”
Keywords
AI regulation, General Data Protection Regulation (GDPR), right to explanation, transparency/explainability, regulatory perspective
We use cookies on our website to give you the most relevant experience by remembering your preferences and repeat visits. By clicking âAcceptâ, you consent to the use of ALL the cookies.
This website uses cookies to improve your experience while you navigate through the website. Out of these, the cookies that are categorized as necessary are stored on your browser as they are essential for the working of basic functionalities of the website. We also use third-party cookies that help us analyze and understand how you use this website. These cookies will be stored in your browser only with your consent. You also have the option to opt-out of these cookies. But opting out of some of these cookies may affect your browsing experience.
