Unlocking The Power of Medical Imaging - New Date

Agenda:
18:00 – 18:50 – Mingling and Interact and food
18:50 – 19:00 – Eyal Hizmi: Opening Remarks
19:00 – 19:30 – Dor Amram: Data On Demand - AIS Data Driven Insights
19:30 – 20:00 - Noa Cahan: X-ray2CTPA: Generating 3D CTPA scans from 2D X-ray conditioning
20:00 – 20:30 - Gefen Dawidowicz: Image-aware Evaluation of Generated Medical Reports

**Dor Amran, Viz.ai**
Data On Demand - AIS Data Driven Insights
This session will highlight Viz.ai’s cutting-edge use of AI to generate actionable insights into Acute Ischemic Stroke (AIS), leveraging advanced CTA (Computed Tomography Angiography) and CTP (Computed Tomography Perfusion) imaging. The discussion will begin by providing a clinical background on AIS, including its prevalence, symptoms, diagnostic challenges, and treatment options. Attendees will learn about the comprehensive suspected AIS patients data collection process, which involves acquiring and analyzing large volumes of patient imaging data. The session will then explore Viz.ai’s multi-model custom-built algorithmic pipeline, specifically tailored to extract critical information such as suspected AIS presence, location, and severity from CTA and CTP scans. Beyond immediate clinical applications, these AI-powered care coordination insights can be harnessed by hospital systems to streamline stroke care pathways, improving resource allocation and optimizing patient outcomes. Public health organizations might leverage it to design targeted prevention programs. Moreover, clinical researchers can use the data to better understand suspected AIS subtypes and trends, driving innovation in stroke therapies and treatment protocols. This session will conclude with a demonstration of how these insights are compiled into detailed data reports, turning patient-specific findings into broader, actionable intelligence to support healthcare innovation and improve stroke care on a larger scale.
Dor Amran is currently a Principal Algorithm Developer at Viz.ai. Previously, he served as an AI Team Manager and Computer Vision Engineer at Viz.ai and has a background in military service with Unit 8200. He holds a BSc in Biomedical Engineering and an MSc in Electrical Engineering, both from Tel Aviv University.

Gefen Dawidowicz, Technion
Image-aware Evaluation of Generated Medical Reports
Our work proposes a novel evaluation metric for automatic medical report generation from X-ray images, VLScore. It aims to overcome the limitations of existing evaluation methods, which either focus solely on textual similarities, ignoring clinical aspects, or concentrate only on a single clinical aspect, the pathology, neglecting all other factors. The key idea of our metric is to measure the similarity between radiology reports while considering the corresponding image. We demonstrate the benefit of our metric through evaluation on a dataset where radiologists marked errors in pairs of reports, showing notable alignment with radiologists' judgments. In addition, we provide a new dataset for evaluating metrics. This dataset includes well-designed perturbations that distinguish between significant modifications (e.g., removal of a diagnosis) and insignificant ones. It highlights the weaknesses in current evaluation metrics and provides a clear framework for analysis.
Gefen is a PhD student in Electrical and Computer Engineering at the Technion in Ayellet Tal's lab. She recently completed an internship as an Applied Scientist at AWS. Her research focuses on developing multimodal models that integrate text and images, specifically in the context of medical and healthcare applications. Her research has been published in top conferences such as ICCV, ECCV, and NeurIPS. She holds a BSc in Biomedical Engineering and an MSc in Electrical and Computer Engineering from the Technion, both with Dean's Honors.

Noa Cahan, Tel Aviv University
X-ray2CTPA: Generating 3D CTPA scans from 2D X-ray conditioning
How nice would it be if patients could undergo a simple, available, and cheap test with low radiation instead of an expensive, high-radiation, and less available test? We developed a generative model that converts 2D chest X-ray images into 3D lung CT scans. We showed that with the generated scans, we can find pathologies that are not visible in X-ray scans and improve the classification of these pathologies from 69% to 80% AUC!
Noa Cahan is a PhD candidate in Electrical Engineering at Tel Aviv University, advised by Prof. Hayit Greenspan. She holds both a BSc and an MSc in Electrical Engineering from Tel Aviv University. Noa's research focuses on deep learning and computer vision in medical imaging, with a particular interest in integrating diverse data modalities such as imaging, free text, and structured tabular data for medical prognosis, as well as developing cross-modal translation models using generative AI.
Noa has been awarded ISF research grants, and her work has been published in leading journals such as Scientific Reports and CMPB, and presented at top conferences including ISBI and NeurIPS. Prior to her PhD, Noa worked at Amazon and Qualcomm.