Teaching AIs to speak the language of hospital services
Hospitals provide essential services for patients, their families and the community, by contributing to their overall health and quality of life. Through innovations and continuous operational improvements the hospitals aim to provide first-rate and cost efficient methods and procedures for diagnostics and treatment.
This project will make it easier to test and evaluate technological innovations based on machine learning and artificial intelligence (AI) for imaging-based examinations in hospitals. Examinations based on CT, MR and PET are already among the most important and most frequently used methods for diagnostics and follow-up of disease or injury. Steady progress in imaging technology has increased the number and complexity of images generated by such examinations, while the number of physicians trained in reading the images has remained the same. Machine learning and AI can provide important tools for radiologists and clinicians, assisting diagnostic processes and patient treatment.
Our project aims to accelerate the translation of new methods from machine learning and artificial intelligence into the hospital, contributing to easier and more precise patient diagnostic workflows. This is done by creating a framework for testing and developing advanced analytical methods by directly linking the algorithms to established hospital workflows, making it possible for hospital personnel to safely and efficiently evaluate new technology, establish a path for rapid prototyping, provide valuable feedback to algorithm developers, and develop trust and expertise in new tools for analyzing medical imaging data.
Recent years have seen exciting applications of machine learning in medical data analysis, from radiology and dermatology to electronic health records and drug discovery. This has led to great interest and enormous expectations from the medical profession. However, it is still early days for the evaluation and integration of artificial intelligence and machine learning-derived information in clinical practice. This proposal addresses one of the crucial missing elements required for implementation and integration in clinical radiology: an innovative, direct integration of computational imaging methods with picture archive and communication systems (PACS). To ensure the usefulness of our innovation, we will apply it to two of the most important health challenges in modern society: cancers in women (uterus) and men (prostate).
We aim to
i) Integrate routine radiology workflow with research infrastructure for computational image analysis.
ii) Apply and evaluate this enabling technology on a large collection of imaging and clinical data from a cohort of 600 gynecologic cancer patients treated at Haukeland University Hospital since 2009. This will be done using novel, deep-learning based approaches to segmentation for ROI-based information extraction ("radiomics").
iii) Investigate how well our innovation generalizes to other computational medical imaging methods and to different applications, and explore how our framework can best be embedded in clinical PACS and HIS/RIS systems.
Such an ambitious endeavor is particularly timely and feasible due to the recently established multidisciplinary center MMIV at Haukeland University Hospital. At MMIV, the project will be embedded in a research environment with access to world class MRI scanners and imaging infrastructure, large collections of patient data, and top expertise within preclinical and clinical disciplines, radiology, pathology, MR physics, mathematics and computational science.
HELSEVEL-H-Gode og effektive helse-, omsorgs- og velferdstjenester