Current Applications of AI in Radiology and its Future Development

The past five years marked a revival of research and applications of Artificial Intelligence in Radiology. A “PubMed” study reveals that in 2018 only 500 annual manuscripts included the terms “Artificial Intelligence” and “Radiology”. In 2020, that number exceeded 2000. As a result of this extensive research and development, novel AI algorithms have the ability to boost radiologists’ efficiency, reduce workflow and increase diagnostic confidence, thus allowing radiology practitioners to focus on their most pressing priorities. Currently, AI is positioned to transform the work of radiologists in three major steps when it comes to image analysis: detection, characterisation and monitoring. [1]

Detection >
Characterisation >
Monitoring >
Key Challenges and the Future of AI in Radiology >

Detection

Detection is the process of identifying and bounding a specific subregion, likely containing a lesion or abnormality. [1] This AI pillar is key to improving workflow efficiency and productivity. Radiology imaging data is growing in disproportionated rates when compared to the number of trained radiology readers. It is estimated that in some cases, a trained reader has to analyze an image every 3-4 seconds for 8 hours in order to catch up with all excessive workload [2]. To achieve better workflow efficiency, radiologists are adopting the usage of AI-based CADe (Computer Aided Detection), utilising a pattern-recognition algorithm. This allows evaluation of larger imaging data volumes in a much faster manner. [1] It’s worth mentioning that these algorithms are not excluding the need of human intervention. Their purpose is to streamline and automate time-consuming tasks and to shift the radiologist’s attention to the image area that matters the most for the patient at hand.

Characterisation

The second AI pillar describes the process of identifying specific qualities of pathologic findings such as size, extent and internal texture. These are used to classify lesions in different diagnostic categories (benign vs. malignant). [1] Leading challenge for radiology readers is the similarity between benign and malignant nodules. This often makes it difficult for readers to accurately diagnose a finding, which also leads to an increased percentage of false-positive and false-negative results, significantly impacting patient outcomes and satisfaction. AI is well positioned to overcome these challenges through its ability to digest and interpret large volumes of imaging data in a reproducible and timely manner, allowing the radiologists to spend more time with the patient, thus improving patient outcomes and satisfaction [1] [4].

Monitoring

Monitoring stands for the follow up of an identified pathology over time, assessing any occurring changes as response to treatment. Often, small changes in lesions are undetectable by the naked eye, and this is where AI algorithms come into play. AI-based monitoring is assisting radiologists by capturing large volume of discriminative features that would otherwise be missed by the radiologist. This allows the AI-based monitoring system to provide a clear picture of the tumour evolution over time. [1] Being able to detect tumor changes in a very early stage significantly increases the patient survival rate and drastically decreases the false-negative result of the examination.

Key Challenges and the Future of AI in Radiology

The benefits of utilising AI-based algorithms in radiology are many and new applications are explored daily. However, there are various challenges that the industry has to tackle before AI can be incorporated in clinical practice at scale. One of the leading challenges is Data Availability. In order for the AI algorithms to work properly, they need enormous data sets, which are still not widely available. Radiology imaging data is spread all across the healthcare continuum, making it difficult to collect and prepare to feed the algorithms [5]. Besides data availability, a pressing challenge for radiologists is developing training and education on AI and how it can be used in their daily practice. For the AI solutions to be implemented and utilised properly, radiologists need to have access to ongoing training education so they can remain up to date with the latest developments in the rapidly developing field. [5]

In conclusion, the possibilities for AI implementation in radiology are numerous and exciting. With many challenges to tackle, the industry will continue developing rapidly and we can expect to see an increased adoption of AI technologies, with various and novel applications. Radiologists will reap the benefits of workflow optimisation by spending more time with patients and less time analysing images. AI algorithms are here to stay and to support radiologists in their most important tasks: providing optimal patient care and improving treatment outcomes in one of the most challenging medical fields.