Chest radiography forms a core component when assessing the degree of lung infection in patients hospitalized with suspected pneumonia as well as the severity of pulmonary edema in patients admitted to hospital with congestive heart failure. Similarly, functional MRIs are frequently employed to monitor the worsening of symptoms in Alzheimer’s patients.

However, assessing medical images is both a time consuming and highly biased process, often requiring multiple experts to review a single image before an agreement on a diagnosis can be reached. This, in turn, acts to delay adequate patient treatment leading to prolonged discomfort and a more laborious recovery period.

As such, adoption of Artificial Intelligence (AI) to assist with medical image analysis continues to garner significant traction. Countless research and development initiatives around the world are focused on creating intelligent solutions that ease the burden faced by clinical professionals. Research on AI technology applied to medical imaging has shown its promise when tasked with identifying infected areas in an image as well as assessing the degree of the infection. A recent study, published in Lancet Oncology, has shown that AI supported screening performed comparatively to standard methodology employed in mammography screening.

Even though AI has the potential to revolutionize healthcare, the adoption of AI for medical image analysis is facing a multitude of hurdles. Research often focuses on ‘beating the rest’ by promoting metrics that have little clinical relevance, limiting a trusted translation to a real-life setting. Further focus on exploring state-of-the-art methodology leads to a lack of insight interpretability, which is critical in settings where how an outcome was reached is equally as important as the outcome itself. Furthermore, many clinical settings use outdated technologies, limiting the ability to integrate complex solutions (either due to the computational size of the solutions or incompatible software present in those settings). This, coupled with data labeling bias and unclear regulatory guidelines, means that the adoption of AI for medical image analysis within a real-world setting is still a while away.

Quantori is looking to refine the adoption of AI for medical image analysis by developing solutions that are both interpretable and capable of achieving performance levels comparable to state-of-the-art models, whilst being modest in their computational size. By working hand in hand with clinicians & radiologists, Quantori aims to understand and address the biggest weak points of AI adoption within healthcare. In the coming weeks we will be sharing a deeper look into how AI can be used to assess the severity of pulmonary edema present in patients’ radiological imaging.