Expert comments

Comments on this technology were invited from clinical experts working in the field and relevant patient organisations. The comments received are individual opinions and do not represent NICE's view.

All 3 experts were familiar with or had used AI technologies before, but not necessarily those in the scope of this briefing.

Level of innovation

All experts considered the technologies to be novel or innovative. One indicated that AI is not currently used in standard care, and another highlighted that it could change the way breast screening is offered. All experts mentioned that other similar, competing technologies are in development. However, these may not yet be commercially available. One expert expressed a desire for more technologies to be included in the evidence section. They provided publications for these technologies to be included. NICE excluded some technologies based on their published process and methods as described in the technologies section. One expert mentioned related technologies, including abbreviated MRI and automated ultrasound. They also mentioned ongoing research to classify according to risk for breast cancer screening, to refine the screening intervals or tests done.

Potential patient impact

All experts agreed that there were potential benefits for patients, including reducing the number of unnecessary recalls, extra visits, and the anxiety these may cause. One expert felt that variability in the recall threshold between different screening services could be reduced, and care could be better standardised. One mentioned a potential decrease in breast cancer mortality because of earlier detection of screen-detected and interval cancers. Another felt that cancer detection would increase, but that this would not improve false negative rates relating to interval cancers. One expert saw an opportunity to develop algorithms that better detect the highest grade cancers, and another suggested that the need for needle biopsy of low-risk lesions could be reduced. One expert raised the important concern that without AI technologies, the NHS may struggle to continue to give breast screening to the current people who are eligible, because of the diminishing workforce. This could lead to patient harm. One expert also felt that those presenting with symptoms could particularly benefit from AI technologies.

Potential system impact

The overall financial impact of adopting the technologies was unclear. One expert suggested that it could be cheaper, and another expected it to cost roughly the same, or perhaps more. Two experts felt that cost savings could be made if the technologies reduced unnecessary recalls, but this would depend on the technology cost. Two experts suggested changes to the pathway by reducing the number of human readers for each case, or by eliminating a human reader entirely. One of these suggested that this would lead to more efficient use of time, and the other expert commented on reducing time taken to read mammograms. Two experts expected that adoption of AI technologies could reduce the number of people needed in the service to read screening mammograms. Two experts also highlighted shortages in the workforce. One expert expressed concerns that without support, such as that offered by these technologies, the Breast Screening Programme may struggle to continue to offer the service to those eligible. Two experts felt that there would be significant work needed to ensure the computing infrastructure of the National Breast Screening Service, and its Picture Archive and Communication System (PACS) was able to support adoption. One expressed that this will be a major issue. Two experts also felt that if AI technologies were used to help those reading mammograms, and not as an independent reader, specialist training would be needed.

General comments

One expert felt the role of the technology is not yet clear. One expert felt it was additional to current standard care. The other expert agreed with this if it was used as a decision support tool, but thought it could replace current standard care if used to read mammograms independently. Two experts referred to usability issues with computer aided detection, the predecessor to AI. All experts felt that more information was needed around how a human reader would interact with the software. No specific safety or regulatory concerns were identified, but 2 experts anticipated potential medicolegal issues if human readers were replaced and cancers were missed. Two experts suggested a number of areas lacking evidence, including the localisation of cancers, the spectrum of disease detected, and independent evaluation of AI technologies in a screening population. One expert highlighted that studies of US radiologists have very low applicability to the UK, because of their higher recall rate and lower reading volume per year. One expert stressed the importance of making sure that the performance of these technologies does not decline if mammography vendors update the software on their machines.