Five CAIM Fellows advance MedTech through AI

AI-Expert Tandem in Pathology

Detecting lymph node metastases in colorectal cancer patients is challenging and laborious for pathologists. The University of Bern's Institute of Tissue Medicine and Pathology has developed an AI algorithm “MetAssist” which already makes accurate predictions but needs refinement to become a robust screening tool to aid pathologists in daily clinical care. Amjad Khan, a postdoctoral researcher at the Digital Pathology lab, will improve MetAssist’s performance by integrating text feedback from pathologists with image predictions using vision-language models – thus combining the strengths of machine performance and human expertise.

Rapid and Complete Tumor Removal

When performing pancreatic surgery it is challenging for the surgeon to accurately assess tumor margins. The current procedure involves sending tissue samples to the pathology department during surgery, waiting for their analysis, and then removing more tissue if needed. A polarimetry-based system, PolaSight, developed at the ARTORG Center for Biomedical Engineering Research, proposes to address this difficulty by analyzing tissue in real time, helping to ensure complete tumor removal and potentially increasing surgerical success rates. Supported by artificial intelligence, the project by Senior scientist Pablo Márquez-Neila is inspired by space technology and performed in collaboration with the Institute of Tissue Medicine and Pathology and the Center for Space and Habitability, University of Bern.

Vision Correction with Deep Learning

By 2050, more than 5 billion people will be affected by various refractive conditions that impair vision. Not all patients will be able to undergo laser surgery to restore their vision. Dr. Miguel Ariza of the ARTORG Center proposes the development of a platform technology, kerAccurate, to correct refractive errors through corneal augmentation. The technology aims to provide a precision treatment that will both minimize complications and optimize vision restoration results. The project is being carried out with ophthalmologists from the Inselspital Eye Clinic.

3D Cardiac Image Reconstruction within Seconds 

Currently, 2D MRI is the main method used to image the structure and function of the heart. 3D MRI methods exist, but they require significantly longer scan times. Scans can be accelerated, but at the cost of longer image reconstruction times. Dr. Eva Peper, PostDoc at the Quantitative MR Imaging Science Lab of the University of Bern and the Inselspital, proposes to use neural networks to speed up image reconstruction times from hours to seconds. This AI-powered approach promises fast 3D imaging, enhancing diagnostic accuracy and patient comfort in cardiac imaging. The project is conducted in collaboration with the Translational Imaging Center (TIC) at sitem-insel.

Informed Hemodynamic Treatment during Anaesthesia

General anaesthesia is performed in millions of patients each year and is considered safe. Yet, a major challenge for the anaesthesiologist is to avoid phases of low blood pressure (hypotension), which may result in poor outcomes, e.g., heart attack or acute kidney injury. Dr. Markus Huber from the Department of Anaesthesiology and Pain Medicine of the Inselspital, aims to unravel the cause-effect relationship underlying hemodynamic management, hypotension and clinical outcomes, using causal inference and machine learning.

(This news has been generated with inspiration by MS Copilot and ChatGPT 3.5)