Imagine there is no paperwork… it’s easy if you try

Abstract

With the current hype of Artificial Intelligence (AI) in healthcare, a classic question arises from the radiologist’s perspective: will radiologists be replaced by AI? In our opinion, the answer is quite simple: just no. In the words of John Lennon (Liverpool, 1940-New York, 1980): “We can imagine there’s no countries”, but we cannot imagine a world without radiologists, since the complexity of their work will be hardly replaced by machines. Currently, there are hundreds of approved AI-based applications for clinical use that try to replace specific radiologists’ tasks, or even, go beyond their diagnostic performance, which could be seen as a threat.¹ Beyond viewing AI as a threat or not, several factors underscore the irreplaceability of radiologists in the short, medium, and even long term. Apart from issues related to diagnostic or interventional procedures, as well as coordinating radiology departments or healthcare resources and human interaction with patients and colleagues, radiologists possess invaluable capability and versatility. They adeptly handle various clinical scenarios, integrating information from multiple sources, not just images or text, but also their expertise and continuous interaction with patients and other clinicians.² These factors, coupled with the inherent limitations of AI tools—often trained for narrow or single-task uses where success in one task does not guarantee success in a similar one—are severely constrained by the quality and volume of data used for training, including costs related to AI solutions development, the “black box” effect, and the challenges of global standardization and external validation of in-house algorithms, keep radiologists far from being replaced.³

However, there are also other kinds of applications that may alleviate the radiologists’ workload, helping (rather than replacing) them in their workflow.⁴ In this manner, imagine a workday without the burden of paperwork, allowing radiologists to concentrate solely on their primary duty: analysing images to provide the most accurate diagnosis, I wonder if it is possible. Moreover, this issue extends beyond just saving time or maintaining their focus on diagnostic duties. It can be part of the solution to the growing global shortage of radiologists, a longstanding yet unresolved problem, and the burnout among radiologists, a growing concern for all radiology departments.^5,6

We believe there are several critical junctures in the radiology department workflow that might benefit from the use of AI tools, as they have the potential to reduce many of the non-interpretative tasks associated with a radiologist’s daily work. Both text-based and image-based AI solutions can be applied at different points in the regular workflow imagers. Among text-based AI algorithms, Natural Language Processing (NLP) tools have widely demonstrated their usefulness in different scenarios.⁷ Tasks related to automatic studies protocolling, scheduling duties, clinical indications and appropriateness criteria for radiological studies, early notification of incidental or relevant findings, improved and efficient communication of radiology reports to clinicians and patients, patients follow-up, recommendation of further exams in radiology reports, or even facilitating billing tasks, are just a few of the potential applications of NLP, currently being explored to mitigate the time-consuming tasks radiologists and rest of staff of radiology department face daily.⁸ Even speech recognition tools employed worldwide, which are also based on NLP algorithms, serve the dual purpose of transcribing and assisting radiologists in real-time dictation, effectively improving the speed of the reporting process. NLP tools also show promise in generating automatic structured reports or crafting conclusions that make radiology reports more patient-friendly.⁹ Among NLP applications, Large Language Models (LLMs) like Generative Pre-training Transformers are witnessing unprecedented attention with promising solutions in various healthcare domains, including radiology. Tools like ChatGPT might help radiologists save time by summarizing electronic health records, structuring radiology reports from free-text dictations, summarizing the conclusions of radiology reports or making them more understandable for patients.¹⁰ However, it is crucial to note that most LLMs have not been trained with specific radiology data. Therefore, radiologists must always review and verify the content derived from these models.¹¹

Yet, there is a subset of AI applications designed to facilitate the radiologists' diagnostic process without directly impacting their diagnostic decision-making. For example, there are AI tools that can automatically perform measurements at radiological studies, primarily associated with cardiovascular or musculoskeletal systems. In terms of facilitating and streamlining the report generation process, AI tools (mostly based on X-ray films) are being developed from a hybrid perspective to extract relevant information from medical images and patient data (eg, electronic health records) to generate preliminary reports.¹²

These routine tasks are usually tedious and time-consuming, and most of them are currently standardized worldwide. Clinical decision support systems, and automatic detection algorithms, including screening tools such as breast imaging, may also facilitate the assessment of radiology images. However, it is always recommendable that radiologists validate or confirm these lesions detected by the AI, serving as a support system for the radiologist.¹³ At this stage, an unavoidable question arises: who is ultimately responsible for the diagnosis, and for potential errors made by AI? This topic is currently a subject of intense debate, as it involves all professionals engaged in the creation and use of AI tools in radiology. There is a real need for radiologists to become familiar with basic AI concepts, and ideally, to be trained in using AI solutions. Responsibility extends to the use of AI tools, and there is a duty to involve radiologists in multidisciplinary teams. These teams are dedicated to developing AI solutions in clinical settings that are transparent and clinically explainable, countering the “black-box” effect often associated with AI. They focus on the use of unbiased and reliable data, ensuring that AI decision-making processes are clear and understandable. These considerations are critical to building trust and accountability in healthcare AI applications and include the need for internal and external validation, quality assurance, and addressing legal issues related to informed consent.^14,15 Informed consent in the context of AI in healthcare involves ensuring that patients are fully aware of and understand how AI is used in their care, the data it utilizes, and the implications of its use. This transparency is essential for ethical practice and patient trust. At this point, stringent regulatory policies are needed to establish legal frameworks and clear guidelines defining AI tools as either merely clinical decision support solutions or as autonomous entities within radiology departments.¹⁶ The legal and ethical implications of these distinctions are significant, and this approach ensures patient safety, ethical use of AI and compliance with legal regulations.

Other AI-based applications are currently in development and undergoing evaluation to support in all these other “non-interpretative” within radiology departments. These tasks include the use of deep learning algorithms at the image acquisition and reconstruction level to improve image quality, reduce acquisition time, and minimize radiation dose.¹⁷ In this realm, AI-based algorithms are being used to decrease the time needed to position and centre patients for MRI or CT scans, also increasing the patients’ security. Also, AI-based tools can reduce the time required to planning, capture, check the quality of acquired images and to automatize postprocessing and quantification tasks, increasing reproducibility of the studies and their accuracy. This time efficiency is vital to optimize the radiology department’s scheduling and workload task.

These AI solutions save radiologist’s time and undoubtedly open a new horizon in the management of radiology departments’ duties by automatizing as much as possible administrative or tedious pre or postprocessing related tasks. Thereby, radiologists are released to focus on improving their diagnostic, teaching and learning skills, including improving communication with patients and clinical interactions with other colleagues. As John Lennon said: “You might say I am a dreamer…”, but this alternative way of saving time and human resources might be perceived as less threatening by the radiological community, especially when compared to tools attempting to supplant the core responsibility of clinical diagnosis.

Funding

This article was partially funded by the Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033), grant number PTQ2021-012120.

Conflicts of interest

A.L., MD, PhD is occasional lecturer of Philips, Siemens Healthineers, Bracco and Canon and receives royalties as book editor from Springer-Verlag.

References