Deep Radiomics Decision Support System for Prostate Cancer Management

Non-invasive medical imaging, especially MRI, has become indispensable in radiology because it enables the assessment of whole organs using different tissue properties to provide quantitative, multi-dimensional and -parametric image data. Clinical evaluation and interpretation of radiological images therefore constitute important part of radiology. Currently, this is mainly done manually and qualitatively by experienced radiologists according to standardized guidelines. Although this has translated into improved healthcare, a major bottleneck in clinical practice is that manual evaluation and interpretation of images for clinical decision-making is cumbersome and subjective, which sometimes lead to under- or over-diagnosis. Importantly, with increasing aging population and accessibility of diagnostic imaging, the number of imaging data continues to increase exponentially. Radiologists are, however, time-limited, and cost-intensive resources that cannot be scaled to future imaging and analysis demands. There is therefore an unmet clinical need for optimized and automated radiological workflow that is reproducible, objective, scalable, and capable of exploiting the full spectrum of clinical information embedded in images for improved patient care. This has triggered, clinical, and commercial interest in AI decision support systems for radiological applications, especially prostate cancer management. Currently, very few solutions exist in this regard, most of which provide partial support and/or are based solely on methods (deep learning) that lack explainability and transparency. The CIMORe group at ISB, NTNU has recently developed a deep radiomics based AI decision support tool that offers performance and explainability for prostate cancer management. The aim of this follow-up project is to investigate the market and business models, and technicalities for delivering the technology with focus on regulation, cost-effectiveness, interoperability, and user-friendliness.

-