2693After specialising in oncology imaging at Manchester, Neelam Dugar is now a consultant radiologist at Doncaster and Bassetlaw NHS Trust and also Chair of the Radiology Informatics Committee at the Royal College of Radiologists. Hospital Healthcare Europe had the pleasure of speaking with her about her career to date and to discuss recent documents produced by the College on the use of artificial intelligence in radiology and the development of a vetting procedure for inappropriate scan requests.
Organisation of services at the Trust
Dr Dugar’s department has approximately 16 radiologists and, with a newly purchased third CT scanner, the department is extremely busy, operating on an almost conveyor belt-like basis. In a typical day, radiographers will perform around 1000 imaging investigations including those for elective, acute and ward patients. As there is a requirement to have A&E CT scans reported on within an hour, Dr Dugar emphasised the importance of prioritising the workload within the team. Hence, one radiologist is always allocated to report emergency scans, and while others do elective work. During the evening and weekends only one radiologist is available for undertaking emergency CT and MRI scan reporting. Night-time emergency radiology has now been outsourced to Australia. Since starting her role over 17 years ago, the workload had expanded considerably due to a combination of increased expectations and national guidelines that often recommend the use of imaging. For example, she described how 17 years ago, during a typical Sunday, she might be called upon to report one emergency scan but on a recent weekend shift, her hospital performed 100 emergency CT and MRI scans. She feels that no other specialty has experienced that kind of explosion in workload.
On the work of the RCR Informatics Committee and her role as Chair
Dr Dugar emphasises that technology is the backbone of radiology and had a desire to make best use of technological advances as a means of enhancing patient care. She explained that she was appointed as informatics advisor in 2015, because of an interest in the topic coupled with the fact that she had led the development of digitisation in her own department. She suggested that the Royal College of Radiologists (RCR) felt that it was necessary to have a committee that was able to set the standards of what should be achieved by all radiology departments implementing informatics. In other words, the RCR Informatics Committee was tasked with defining the standards and hence, best practice, which should be achievable through hospitals’ information technology (IT) systems.
Out of this work came the publication Integrating artificial intelligence (AI) with the radiology reporting workflows (see later in the supplement for a summary of guidance). The guidance defined the standards for how AI should be incorporated into the radiology information (RIS) and picture archiving and communication systems (PACS). Dr Dugar highlighted how, in some respects, AI is considered a very broad term and can be interpreted differently depending on the context. For the present document, the RCR informatics group considered AI in the narrow context of ‘computer vision’ used for radiology image pre-analysis. As Dr Dugar explained, because AI systems have already been developed for facial recognition, given that the role of a radiologist is to visualise images and to make interpretations to inform the ongoing care of patients, it seemed only right that this should be the area to focus upon.
However, a primary focus was to ensure that IT vendors could develop the necessary infrastructure to incorporate AI systems with different hospitals. A further consideration for the implementation of AI was the apparent national shortage of radiologists in the UK. For instance, in a 2018 report,1 it was noted how in the UK, only one in five UK Trusts and health boards had enough interventional radiologists to provide safe 24/7 services to perform urgent procedures.
Dr Dugar defined how the AI workflow guidance should work in practice, explaining that the AI system would initially review the image before it was seen by a radiologist. The AI system algorithms were such that it was able to highlight any relevant features, which is also within the remit of radiologist. Nevertheless, whereas the AI systems are capable of detecting some abnormalities, the radiologist would then combine these findings with other test/imaging results, and any other relevant clinical findings, to create a more personalised report for the patient.
Will AI replace radiologists in the future?
If the AI system can do the essential job of a radiologist, surely these individuals can be easily replaced? Dr Dugar disagrees. She revealed how in 2016 AI pioneer, Geoffrey Hinton, had said “we should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists.”2 At the time, she said this created a major staffing crisis, especially in the US, which saw a downturn in doctors choosing radiology as a career, fearing that they would be deemed superfluous in the near future. But, as Dr Dugar added, medicine is not maths – if it were then there would be no need for a radiologist! She feels that it is virtually impossible to create an ‘artificial’ radiologist simply because of the huge number of algorithms that would be required to emulate the thought-processing of a radiologist.
Dr Dugar believes that having an AI system evaluating a scan goes some way towards the need for two independent reviewers of a scan result. This is considered as best practice, lending support to the metaphor that “two heads are always better than one”. As she said, this is the current recommendation for breast cancer screening. She stressed that having two independent reviews was necessary because even though individual radiologists are highly trained, they are fallible. Thus, from a safety perspective, dual review is the ideal standard.
The integration of an AI system was also important but for a very different reason. Dr Dugar highlighted that, in reality and with a national shortage of radiologists, it becomes impossible to achieve the two-reporter standard for images. One of the key reasons for a second reporter was to the minimise the phenomenon of satisfaction of search, which describes the situation where some lesions remain undetected after an initial lesion. As Dr Dugar illustrated, when a radiologist finds an abnormality, they become fixated on that particular problem and start to process this finding within the context of other clinical information and sometimes ignore other findings. With an AI system able to review the image prior to the radiologist, it effectively becomes that second reporter and a helper, alerting the radiologists to the full range of abnormalities present on the image. In discussion with colleagues, a barrier to greater use of AI is the perception among some radiologists that the system is very sensitive but not specific. Using the example of the assessment of a lung scan, Dr Dugar explained that while the AI system would report on the presence of tiny nodules, the focus of the radiologist was in looking for metastases. With greater knowledge of the patient’s clinical history than the AI system, the radiologist can potentially discount the relevance of these nodules and advise accordingly. In contrast, the AI system simply identifies any abnormality and is unable to make a subjective judgement within the context of any other clinical findings. Dr Dugar labelled the AI system as a ‘junior radiologist’, i.e., it was able to provide a limited role as a preliminary reviewer on images. Nonetheless, she did believe that in the future, with improvements in AI occurring, the input from a radiologist might be unnecessary, especially for simple imaging, e.g., reporting on the presence/absence of a fracture. However, radiologists would still be required to interpret more complex imaging from MRI or CT scans. She added that while an AI system could identify a filling defect in the lungs and report the most likely cause to be a pulmonary embolism, as a radiologist, you are always thinking laterally about other differential diagnoses.
On the vetting and cancellation of inappropriate scan guidelines
As Dr Dugar described, being both medically qualified and trained in radiology allowed her and her colleagues to assess whether or not a particular imaging request was appropriate. She emphasised how often both junior doctors and those from other specialities, may not be completely clear on which imaging tests were correct. The vetting (triaging) and cancellation of inappropriate radiology requests document was introduced simply to help manage the workload within radiology departments. An important part of Dr Dugar’s role is to always vet or triage any requests for imaging that reach the department. This vetting process, she added, was crucial because of the high workload of the department, which makes it impossible to perform every exam request.
While the vetting process amounts to a clinical assessment task in itself, Dr Dugar highlighted how within her department over 90% of the vetting process was undertaken by the radiographers rather than the radiologists. This had been made possible through the introduction of a protocolised approach for the radiographers. Moreover, Dr Dugar believes that radiographers can quickly acquire the necessary vetting skills and then approve and book a scan or reject the request. A further advantage to radiographer-based vetting, is that, as these individuals are involved in performing the imaging, they are able to quickly assess, and then cancel, any duplicate requests and in some cases, even determine if the request would be of additional value, i.e., if the request is for a broadly similar scan. A difficulty for radiographers, however, is that without the necessary medical training, they may feel uncomfortable cancelling an imaging request that was requested on clinical grounds and in such instances, the protocol would dictate that the request is forwarded to the radiologist. As a consequence of introducing the vetting process in her department, Dr Dugar felt that on a typical day, she might be asked to vet up to 30 requests and allocates up to 30 minutes of her day to this task. She thinks that such vetting is a key task given that the department performs around 1000 scans each day. Although some of the requests passed to her from radiographers can be challenging, in many cases, it can sometimes be very straightforward and require simply altering the request to a more appropriate imaging modality. For more complex cases, she will need to review the patient’s medical history or initiate a discussion with the referring clinician to discuss the best option. In cases where the request is rejected, Dr Dugar ensures that the requesting clinician is informed of her decision and the rationale behind the cancellation. She thinks that the departmental system, being fully electronic has streamlined the whole request/cancellation process.
Dr Dugar expressed the view that the vetting document was desperately needed because in some radiology departments, no vetting process was in place. She realised that part of the reason behind this lack of vetting was largely due to a lack of functionality within the hospital’s internal IT system. The purpose of the vetting guidance was thus to ensure that while not all NHS Trusts employed the same vendors, these vendors would modify the IT infrastructure to enable electronic communication for the vetting process. As Dr Dugar said, in discussion with radiologists from outside of her own department, the cancellation process was often not communicated to the original requesting clinician and this led to internal friction and, in some cases, the radiologists in an attempt to appease the requesting clinicians, decided to no longer triage requests, with a resultant increase in their workload. Although it seems unusual that the whole of the NHS must deal with different IT vendors, Dr Dugar is against the idea of a national vendor. She thinks that with such a huge monopoly, there would be little incentive to innovate. What is more important she feels, is that the same workflow processes should be adopted in the different NHS Trusts, to improve the efficiency of radiology departments, even if this occurs through dissimilar IT systems.
On the impact of the pandemic on imaging services
Dr Dugar said that for her, radiology services did not stop during the pandemic although the focus shifted to COVID patients. She felt that her own work, which is either cancer or emergency-based, did not slow during the pandemic. She believed that one of the greatest changes because of the pandemic was the digital transformation within the NHS. She thinks this was of enormous benefit, enabling more virtual meetings which were a great advance compared to teleconferences. Another important development for work–life balance was allowing radiologists to have workstations at home. Dr Dugar says that having an interest in digital technology, she had tried to implement greater home working for some time, but her request was always denied due to lack of funding. However, she also thinks in the future, this innovation of homeworking and virtual meetings will not revert to pre-pandemic times but there will still be a balanced need for office working.
On the evolution of the imaging landscape over the next few years
Dr Dugar believes that AI algorithms will develop in the next two to five years and become a much better preliminary reporter on many more things such as fracture detection, lung nodule detection etc. She mentioned how AI is already being used in brain imaging for strokes. She worries, however, that future innovations in AI by computer scientists will require additional funding, and that this should not be at the expense of a radiologist training.
Another potential growth area she feels is in the evolution of enterprise imaging3 and revealed how all her own radiology department’s images have already been archived and made available throughout the enterprise. An important current problem, she explained, was how various medical images from other specialties/departments have been generated but are stored in different locations and formats without the correct patient identifiers etc. and are not indexed properly (e.g., endoscopy images, ECG, audiometry, sleep studies etc). Incorporation of all the images and graphs in a single and accessible location will be of enormous value, not only to radiologists but also all treating physicians. With the ability to review all images, and together with other pieces of clinical data, it will allow radiologists to create a much more personalised report for the patient.
Although improvements in smartphone technology allow for image review, Dr Dugar thinks that at the present time, the quality of the images is not of sufficient quality for diagnostic purposes. Furthermore, from a medico-legal perspective, she would not use the images reviewed on a smartphone for reporting. She also felt that while mobile scanning units for MRI and CT scans were available and could be utilised for elective work, these imaging modalities would still need to remain within the hospital premises, where the equipment was needed for emergency scans.
Image acquisition and interpretation were separate, and Dr Dugar believes that she does not need to be present at a mobile scanning unit and can remain either in the office or at home to undertake her interpretive role for the images. However, radiologists (whether remote or on-site) must continue to work closely with the radiographers operating the scanners to provide support and advise on appropriateness and vetting/triaging support. Radiologists and radiographers must always work as teams to improve patient care.
References
- Bassett M. Radiologist shortage deepens in the UK. www.rsna.org/news/2019/May/uk-radiology-shortage (accessed June 2021).
- Marcus G, Little M. Advancing AI in health care: It’s all about trust. www.statnews.com/2019/10/23/advancing-ai-health-care-trust/ (accessed June 2021).
- Petersilge C. The evolution of enterprise imaging and the role of the radiologist in the new world. www.ajronline.org/doi/full/10.2214/AJR.17.17949 (accessed June 2021).
