A series of studies have described the use of deep learning algorithms to detect abnormalities in radiology, yielding promising results. ‘qXR’, Qure.ai’s chest X-ray interpretation tool, is able to automatically detect and localize up to 29 abnormalities, including those indicative of possible lung cancer. There are several features of chest radiographs (such as sharply circumscribed nodules or masses, those with irregular margins, and those with ill-defined lesions) that can indicate the presence of lung cancer. The CE-marked qXR algorithm can not only detect lung cancer nodules with high levels of accuracy in under a minute, but also marks out the position and size of these nodules. It can aid clinicians in picking up minuscule nodules which may be missed even by experts. A study conducted by Qure.ai demonstrated a 17% improvement in sensitivity when using AI to interpret chest X-rays, compared to radiologist readings. Such aids in early detection can have considerable long-term benefits for medical professionals in their efforts to tackle lung cancer. It can also mean lower cost per-life-year saved.
Another area where AI can be very helpful in patient management in radiology is teleradiology. Teleradiology is the electronic transmission of radiological patient images from a scanning organization to a different reading organization, for the purposes of diagnostic interpretation and reporting. In 2019 less than 2% of the 4.7B diagnostic imaging scans performed globally fell into this category.
However, several drivers will contribute to this penetration rate increasing significantly over the next five years. These include:
-
Shortages of radiologists in certain countries/regions
-
Increased demand for more specialized modality reads such as CT and MRI, that require radiologists with specific skills
-
Demand for out of hours reporting, particularly in time-critical applications, e.g. neurology
-
Longer read times for more specialized modalities (e.g. CT v X-ray)
-
Increased use of cloud-based technology making implementing IT for teleradiology less complex
-
Changes in legislation that support reading services being provided out of country and by third parties
-
Increasing numbers of imaging IT vendors offering workflow tools that are designed specifically for teleradiology applications
One final driver, not mentioned above, is that of technology advances, in particular those relating to AI.
Three of the most important factors that heavily influence the success of teleradiology reading service providers are:
-
The speed that radiologists working for teleradiology service providers perform and report on their reads.
-
The accuracy of the reports produced by radiologists.
-
The workflow and decision support processes that service providers put in place to ensure urgent reads are prioritized and reported on quickly.
Over time, AI can be used to support and improve all three. The first two bullets (read speed and accuracy) will become increasingly important as the types of diagnostic imaging scans being performed become increasingly focused on more complex and time-consuming procedures. For example, 11.5% of diagnostic imaging procedures performed globally in 2019 were CT scans, this ratio has been increasing steadily over the last decade and is projected to reach 14% of all diagnostic imaging scans in 2024. Conversely, X-rays accounted for 61% of all scans in 2019, a figure that is projected to fall to approximately 55% in 2024.
