Survey with healthcare professionalsResponse rate and participants demographics

A total of 95 HCPs completed the survey. Majority of the respondents were male (60%, n = 57/95). Nearly two third of respondents (62%, n = 59/95) were between ages 35 and 54 years and had more than 10 years of experience (65%, n = 62/95). HCPs from Greece (36%, n = 34/95) and Italy (26%, n = 26/95) constituted the majority of the respondents. Participants’ demographics are presented in Table 2.

Table 2 Characteristics of HCPs who participated in the surveyHealthcare professionals’ experience and perceptions of the four cancer care pathwaysExperience and perceptions of the cancer care pathwaysBreast cancer

About half of the HCPs (53%, n = 50/95) were involved in the care pathway of breast cancer. Participants were asked about the accuracy of the current imaging techniques used in breast cancer care in terms of false positives and false negatives. In this regard, 64% (n = 32/50) reported the occurrence of false positives. According to the respondents’ experience, the imaging techniques associated with false positives in order of frequency were mammography (n = 15), magnetic resonance imaging (MRI) (n = 13), PET/CT scan (n = 7), ultrasound (US) (n = 2), bone scan (n = 2) and computed tomography (CT) scan (n = 1). On the other hand, 50% (n = 25/50) reported the occurrence of false negatives. The imaging modalities associated with false negatives in order of frequency were mammography (n = 11), US (n = 9), PET/CT scan (n = 6), bone scan (n = 2) and MRI (n = 2).

Overall, 78% of HCPs (n = 39/50) perceived the current care pathway to be efficient. Nevertheless, more than half (56%, n = 28/50) indicated that patients face delays in diagnosis. Delay in referring patients to diagnostic tests/images was the main cited reason as indicated by 75% of HCPs (n = 21/28), followed by lack of adequate imaging resources/equipment (14%, n = 4/28), and lack of adequate staffing/understaffing (3.5%, n = 1/28). Furthermore, 40% (n = 20/50) indicated that patients face challenges particularly related to imaging tests, including: long waiting lists and delays in appointments (n = 9), lack of expertise and availability of imaging resources (n = 6), psychological distress due to false positives and false negatives (n = 2), and patients being reluctant to undergo diagnostic tests (n = 2).

Lung cancer

Just in excess of two fifths of the HCPs (44%, n = 42/95) were involved in the care pathway of lung cancer. Participants were also asked about the accuracy of the current imaging techniques for this cancer type in terms of false positives and false negatives. In this regard, 60% (n = 25/42) reported the occurrence of false positives. The imaging modalities associated with false positives in order of frequency were PET/CT scan (n = 16), followed by CT scan (n = 12). On the other hand, 57% (n = 24/42) indicated the occurrence of false negatives. The imaging techniques associated with false negatives in order of frequency were chest x-ray (n = 8), PET/CT scan (n = 5), followed by CT scan (n = 4).

Overall, 67% of HCPs (n = 28/42) perceived the current care pathway to be efficient. Nevertheless, nearly two-third (64%, n = 27/42) indicated that patients face delays in diagnosis. Delay in referring patients to diagnostic tests/images was the main cited reason (78%, n = 21/27), followed by lack of adequate staffing/understaffing (18.5%, n = 5/27). Additionally, 45% (n = 19/42) indicated that patients face challenges particularly related to imaging tests including: long waiting lists and delays in appointments (n = 10), lack of expertise (n = 4), lack of imaging resources (n = 5) and delays in imaging reporting (n = 2).

Colorectal cancer

Just less than half of the HCPs (47%, n = 45/95) were involved in the care pathway of colorectal cancer. Participants were also asked about the accuracy of the current imaging modalities in terms of false positives and false negatives. In this regard, 51% (n = 23/45) reported the occurrence of false positives. The imaging techniques associated with false positives in order of frequency were PET/CT scan (n = 14), CT scan (n = 6) followed by MRI (n = 5). On the other hand, 47% (n = 21/45) indicated the occurrence of false negatives. The imaging techniques associated with false negatives in order of frequency were CT scan (n = 8), MRI (n = 7), PET/CT scan (n = 3), followed by FDG PET/CT scan (n = 2) and colonoscopy (n = 2).

Overall, 56% of HCPs (n = 25/45) perceived the current care pathway to be efficient. More than three-quarters (76%, n = 34/45) indicated that patients face delays in diagnosis. Delay in referring patients to diagnostic tests/images was the main cited reason as indicated by 56% (n = 19/34), followed by lack of adequate imaging resources/equipment (21%, n = 7/34) and lack of adequate staffing/understaffing (15%, n = 5/34). In addition, 40% (n = 18/45) indicated that patients face challenges particularly related to imaging tests including: long waiting lists and delays in appointments (n = 5), lack of imaging resources (n = 5) and lack of expertise (n = 4).

Prostate cancer

Two fifths of the participants (40%, n = 38/95) were involved in the care of prostate cancer. Participants were asked if they have encountered any false positives and false negatives during the diagnostic procedure for prostate cancer. In this regard, 45% (n = 17/38) reported the occurrence of false positives. The imaging techniques/tests associated with false positives in order of frequency were prostate MRI (n = 4), PSMA PET/CT scan (n = 2), followed by choline PET/CT scan (n = 1) and PSA test (n = 1). On the other hand, 61% (n = 23/38) were aware of the occurrence of false negatives. The imaging techniques/tests associated with false negatives in order of frequency were biopsy (n = 4), MRI (n = 3), followed by FDG PET/CT scan (n = 1) and bone scan (n = 1).

Overall, 68% of HCPs (n = 26/38) perceived the current care pathway to be efficient. Less than half (42%, n = 16/38) indicated that patients face delays in diagnosis. Delay in referring patients to diagnostic tests/images was the main cited reason (56%, n = 9/16), followed by lack of adequate imaging resources /equipment (25%, n = 4/16) and lack of adequate staffing/understaffing (6%, n = 1/16). In addition, 39% (n = 15/38) indicated that patients face challenges particularly related to imaging tests, including: long waiting lists and delays in appointments (n = 4), lack of availability of imaging resources (n = 4).

General challenges within the care pathway

HCPs were required to list three main challenges affecting the cancer care pathway, via an open-ended question. A total of 14 challenges were identified by content analysis across the 4 tumour types. The challenges were classified into 4 categories: (i) healthcare system related challenges, (ii) HCPs related challenges, (iii) clinical practice related challenges and (iv) patients related challenges. Across the four tumour types, the leading challenges were health care system related ones with the long waiting lists/times for diagnosis and treatment. The second challenge was lack of the different imaging modalities/tests within the institutions. Whereas in the third rank came the issue of lack of manpower/understaffing. The full list of challenges across the four tumour types is presented in Table 3.

Table 3 General challenges within breast, lung, colorectal and prostate cancer care pathwaysChallenges in cancer care that can be addressed using AI and ML

Using an open-ended question, HCPs were also required to list three main problems/challenges related to the use of imaging that could be resolved using AI and ML. A total of 10 main challenges were identified by content analysis across the 4 tumour types. The challenges were classified into 3 categories: (i) healthcare system related challenges, (ii) HCPs related challenges and (iii) clinical practice related challenges. The leading challenge that HCPs thought would be improved with AI and ML was clinical practice challenge related to the accuracy of the current imaging modalities/tests in terms of reporting and interpretation, and rates of false positive and false negative. The second was health care system related and constituted the reduction of long waiting lists/times for diagnosis and treatment, whereas the third was a clinical practice related challenges with disease evaluation in terms of characterisation, differentiation, and staging. The full list of challenges is presented in Table 4.

Table 4 Challenges that can be solved using AI and ML techniques across the four cancer typesExperience of technology use in oncology practice and acceptance of further technological interventions

All participants (n = 95) agreed that the use of technology would improve the care pathway for cancer patients. However, the majority (73%, n = 69/95) indicated no prior use of technology within the care pathway, versus 27% (n = 26/95) who have used technology and are still using it. Interestingly, all of those who are using technology but one (96%, n = 25/26) indicated that technology use is making a significant difference to patient’s care. The technologies currently used by some of the HCPs include Computer Aided Detection (CAD) and Picture Archiving and Communication System (PACS) systems.

The vast majority of respondents (89%, n = 85/95) indicated their willingness to deliver AI-based services to optimise medical imaging in cancer care in the future. Participants were asked via an open-ended question as to how an AI-based technology can gain the trust of HCPs and facilitate adoption. A total of 6 suggestions were identified by content analysis. The suggestions were classified into two main categories: (i) suggestions related to HCPs and (ii) suggestions related to the technology itself. The main and foremost suggestion was related to having an AI-technology that is evidence-based via randomised controlled trials (RCTs) to support its validity, reliability, and effectiveness. The second suggestion was the need to demonstrate the relative advantage of the AI-technology compared to current practice. The third was the provision of training and education. The fourth was pertinent to raising awareness about the technology among HCPs. Two suggestions were identified in the last rank, one was related to the fact that the AI-technology should not be perceived as replacement to the role of HCPs in decision making, and the other was the ease of use of the technology itself (Table 5).

Table 5 Suggestions provided by HCPs that can facilitate the adoption of an AI-technology and gain the trust of HCPs

Using an open-ended question, participants were also required to elaborate on the elements/characteristics required in an AI tool in clinical practice. A total of six elements were identified. The first and foremost element was having an AI-tool that acts as a clinical decision support tool rather than a replacement for HCPs/medical expertise in clinical decisions (i.e. optimisation of decision making), so cross checking with expert advice needs to be available and the final decision needs to be for the HCPs. In the second rank, accuracy of the tool, and validity were chosen as important elements. Thirdly, ease of use, and reproducibility. Whereas the last element was related to AI explainability (Table 6).

Table 6 Elements an AI tool that would reinforce control in clinical practice as perceived by the participating HCPs

In addition, participants were asked about the best place within the care pathway for introducing an AI-based technology for optimisation of cancer imaging. Screening was chosen as the best place for the introduction of such technology by 40% (n = 37/95) followed by initial diagnosis (36%, n = 34/95) thereafter, further examination, disease staging and differentiation (16%, n = 15/95). Monitoring of treatment was the least favourable location for the introduction of an AI with only 9.5% of responses (n = 9/95).

Interviews with healthcare professionals

Analysis of interviews revealed three main themes with associated subthemes.

Lack of widespread preparedness for AI in oncology

None of the respondents indicated prior or current use of AI in their clinical practice. Many HCPs perceived AI to be still in the initial or infancy phases with respect to oncology care. Hence, HCPs’ perceptions were divided between advocates and sceptics.

“I think we are in the very beginning, and I think we have to check the tool first, and we need experience before saying how confident I would feel about the tool.” (HCP8-Spain)

“…However, we are still in our infancy in the AI technologies.” (HCP10- Cyprus)

Some respondents advocated the introduction of AI via highlighting the potential benefits that AI would bring to oncology practice including: (1) aiding in clinical decision making, (2) promoting the efficiency of cancer care via making processes smoother and thus reducing the time spent across the different stages within the pathway, (3) reduction of interobserver variability, in addition to (4) reduction of clinicians’ workload via making tasks much quicker. The respondents reflected on several time-consuming tasks which they envisaged can be automated using AI such as tumour contouring, image segmentation, image quality checking, cases triaging and prioritisation.

“I think AI technology will play important role in diagnosis as it can speed up the process in diagnosis. Moreover, the AI tool will help doctors to contour the tumours more accurately…. It will help doctors to do their jobs faster, their workload is very high nowadays.” (HCP4- Greece).

“There are many advantages, more accurate diagnosis, significant reduction on time, reduction of delays in treatment and delay of therapy…. Interobserver variability may be reduced by AI…. (HCP7-Italy).

Whereas other HCPs were sceptical about AI introduction. Fear of jobs replacement by AI was raised by some respondents, which was intertwined with the issue of deskilling of HCPs as a result of over-reliance on AI in clinical practice. According to HCPs, losing clinical decision-making skills could lead to overlooking mistakes and errors that AI tools may produce, thus risking patients’ safety.

“…so the medical experience is gone, it would be gone in some years. You know when we extensively use AI and for a lot of medical professions like radiology, radiation oncology and so on. When we have just the AI doing the job, the physicians won't be able to do their jobs anymore of course, this is one problem, but on the other hand, who is checking the AI?’ (HCP18-Cyprus).

“There is a possibility for radiologists to lose their skills and ability to perform. Our capabilities on making difficult diagnoses might be affected because there is an AI that can do it for us.” (HCP5-Italy)

Barriers of AI in cancer care

Several barriers were articulated by HCPs regarding the introduction of AI in oncology care. Cost of implementation and infrastructure were identified as main barriers. Other barriers included: lack of HCPs’ time, age of HCPs as more senior colleagues might not be confident in using AI and could be sceptical as they are used to the more traditional ways of work, ethical issues surrounding data privacy and sharing, accountability in case of disagreement or when things go wrong in practice, lack of training and education in addition to fear of jobs replacement by AI.

“Well, it will be the funding. I think it's important. I think this might be another challenge, for example, to convince the people that they will not be replaced.” (HCP3-Greece)

“I can think of infrastructures right, because I mean, you would need access to this tool, so hardware everywhere, but also overlaps with the cost issue …based on my personal experience, many professional workers in healthcare are quite old. That's a gap that could be really hard to fill…. So, education, but just in point so if we do not want to wait for a generational change, then there’s a lot of teaching to do.” (HCP17- Italy).

“I think the major challenges to its confidentiality and where you're going to store all this data.” (HCP12- Cyprus)

“…to ensure privacy of data and to have legal and ethical clarity. And of course, there is the question if a diagnosis is wrong, who is responsible for this? Is it the AI algorithm, who wrote the algorithm or the doctor? Okay, there are such questions that are difficult to answer.” (HCP16- Greece).

Some respondents also cited patients’ perception of AI as a barrier. From HCPs’ perspective, lack of patients’ awareness about AI as a technology might cause agitation and disbelief, leading to a bad rapport between clinicians and patients. Thus, participants reported that increasing patient and public awareness of the advantages and benefits of AI in clinical practice is crucial for its effective implementation.

'At the beginning of the usage of AI, patients may be sceptical about whether these tools have some negative impact on their health…..' (HCP6-Serbia)

“…. I have read in some articles that patients do not trust an AI tool for their diagnosis…” (HCP16- Greece)

Additionally, the majority of the participants reported that explainability and interpretability are barriers to adopting AI in clinical practice. HCPs were against having an AI tool that functions as a black box. In addition, participants perceived that HCPs will only be able to cooperate with the novel AI technologies when they can understand how AI models work and what factors the AI models use to achieve clinical conclusions/decisions. Participants perceived that lack of transparency in the AI decision-making process could cause dilemmas and confusion among HCPs. Issues related to data availability, data quality and harmonisation also emerged as barriers.

“I will have the feeling of controlling the AI when I have the explanations of the parameters that the AI uses and to understand how AI is working or doing with the images, for example to have clear knowledge about the parameters used for analysis … I do not want to see it as a black box” (HCP4-Greece).

“I think the biggest challenge is to have a big sample of data, and the data to be harmonized and of quality….” (HCP26- Cyprus

Facilitators of AI in cancer care

Education and training of HCPs in addition to raising awareness among patients were depicted as crucial facilitators for AI implementation to alleviate any potential fear associated with the introduction of new technologies.

“…Patient awareness and training of healthcare professionals are important.” (HCP6-Serbia)

“I think there should be a training despite if it's easy, there should be a training because people sometimes are scared about the things they don't know.” (HCP20- Spain)

Having a tool that is both easy to use and user friendly, and evidence-based in terms of accuracy, reliability and validity also came as facilitators. The HCPs highlighted the need for a tool that will not require a lot of time and data input due to their immense workload.

“Reliability. If it's reliable then they will adopt it. If not then they will say I'm better than this so if it's reliable, if it shows that it can produce reliable results, then it will be adopted. Rigorous validation testing so they (referring to AI tools) would gain the trust.” (HCP1- Cyprus).

“…. I think first it should be user-friendly, obviously because everybody is very busy and the technology is advancing on the time, so it should be user friendly. (HCP12- Cyprus)

Time was also envisaged as an essential facilitator for AI implementation, as some respondents perceived that time is needed to allow HCPs to trial the AI tool in their institution and to adjust to experiencing it in order to see how it works and how they would incorporate it into their daily work routines.

“To work with the software for certain period as trial to practically its reliability. Not to be forced to use it. I would suggest internal trial for the software with all the doctors so then the tool can be validated in our clinical practice.” (HCP5- Italy).

“I think that everyone that has the AI tool wants time to play with the AI tool. To see the accuracy of the diagnosis, the sensitivity, the false negatives to see how it works and I think time is crucial for the implementation.” (HCP16- Greece).