Skip to content
Medical Machine Learning

Details

Agenda:

18:00-18:30: Networking
18:30-19:00: Prediction of cancer treatment response from histopathology images through imputed transcriptomics, by Leon Gugel
19:00-19:30: Machine learning solutions to prevent paralysis during spine surgeries, by Einat Kermany, PhD
19:30-20:00: A label efficient approach for structures segmentation in fetal MRI scans, by Bella Specktor Fadida

Prediction of cancer treatment response from histopathology images through imputed transcriptomics
In recent years, the use of tumor molecular profiling in the clinic has allowed for more accurate cancer diagnostics, as well as the delivery of precision oncology. Rapid advances in digital histopathology have allowed the extraction of clinically relevant information embedded in tumor slides by applying machine learning methods, capitalizing on recent advancements in image analysis via deep learning.
In this lecture, we will introduce ENLIGHT-DP, developed by Pangea Biomed. This method is the first generic deep learning based methodology for generating histopathology-based predictions of patients’ response for a broad range of cancer therapies, which importantly does not require matched histopathological slides and response data for training.
Leon Gugel is a senior researcher in the biotech company Pangea Biomed. He has an industry record in classical machine learning, data science, deep learning, generative AI, medical imaging, inverse problems & signal processing. His research interests include applied math with applications to image and signal processing, inverse problems and machine learning.

Machine learning solutions to prevent paralysis during spine surgeries
Spinal surgery presents a significant level of risk, with potentially severe complications such as paralysis and enduring sensory impairment. Fortunately, many of these complications can be prevented or mitigated through the application of Intra-Operative Neuromonitoring (IONM). Although the realm of IONM is relatively young, it is swiftly evolving into a standard of care within neurosurgery, orthopedic, and ENT (ear, nose, and throat) procedures. During the course of neuromonitoring, pertinent biological signals are recorded and analyzed both before and during surgery, allowing neurophysiologists to identify impending neurological issues.
Nervio is a pioneering startup company established to address challenges in this field by providing a machine learning-based solution.
In my presentation, I will share the company's journey, starting with an overview of neurophysiology in general and, more specifically, intra-operative neuro monitoring (IONM). Following that, I will drive into our solution, discussing the complexities stemming from the diverse range of information sources and the high variability of data. I will also outline our strategies for addressing these challenges, including the presentation of some results we have achieved.
Einat Kermany, PhD is a CTO at Nervio Neurophysiological AI. Before that, she was a research staff member at IBM research.

A label efficient approach for structures segmentation in fetal MRI scans
Fetal MRI has the potential to complement US imaging and improve fetal development assessment by providing more accurate volumetric information about the fetal structures. However, volumetric measurements require manual delineation, also called segmentation, of the fetal structures, which is time consuming, annotator-dependent, and error-prone. State-of- the-art automatic segmentation methods for volumetric scans are based on deep neural networks. While effective, these methods require a large, high-quality dataset of expert-validated annotations, which is very difficult to obtain.
In this talk, we introduce a bootstrap segmentation approach that utilizes label-efficient methods to segment structures in fetal MRI scans. We showcase the effectiveness of a TTA-based self-training technique, both with and without active learning, to enhance segmentation performance when dealing with limited data. Moreover, we illustrate how partially annotating scans can enhance the robustness of segmentation. We then demonstrate a clinical application of automatic fetal weight estimation in MRI using a large data cohort constructed with the bootstrapping methodology. In addition to reaching high weight estimation accuracy, repeatability, and reproducibility, it presents a new MRI growth chart and demonstrates the ability to automatically identify FGR fetuses.
Bella specktor Fadida is a PhD candidate at the Hebrew University under the supervision of Prof. Leo Joskowicz. Prior to that, she worked on medical imaging algorithms as a scientist for 7 years at Philips. Involved with the Israeli machine learning community, Bella is also the founder and organizer of the Haifa Machine Learning and Machine Learning for Medical Imaging (MLMI) meetups.

Photo of Haifa Machine Learning group
Haifa Machine Learning
See more events