| Popis: |
BACKGROUND Medical imaging modalities, such as X-rays, ultrasound, magnetic resonance imaging (MRI), computed tomography (CT) and positron emission tomography (PET), have played pivotal roles in the early detection, diagnosis and treatment of diseases in the past few decades (1). Human experts are primarily responsible for the interpretation and analysis of medical images in clinical work. Medical doctors have recently started to benefit from computer-aided diagnoses (2). Advances in machine learning methods have resulted in powerful learning algorithms, collectively referred to as artificial intelligence (AI) (3), defined as computer systems that can perform tasks that would normally require human intelligence, such as language translation, decision making, speech recognition and visual perception (4). In medical fields, AI is widely used, especially for domains that need imaging data analysis, such as diagnostic imaging (5, 6) and pathology (5). In diagnostic fields, such as radiology, this approach carries the potential to transform the clinical practice of physicians. Although the vast majority of applications have been focused on assisting and augmenting radiologists, there is a growing niche of applications that are directly appropriate to radiography practice (6, 7). AI is recognized for providing unique benefits in medical imaging practice, such as reduced workplace-related stress and rates of diagnostic errors and providing clinical decision support to radiographers and radiologists (8, 9). The perspectives of practising radiographers and radiologists on the integration of AI into medical imaging are poorly understood. It is unclear whether AI will support radiographers’ decisions in radiation dose selection. It is mandatory for healthcare professionals to be able to predict the professional requirements and potential unknowns of AI in order to ensure its safe, continuous and effective integration into medical imaging practice (10, 11). This has sparked debates about the responsibilities and roles of imaging professionals, especially radiologists and radiographers, who will use this technology (10). During the COVID-19 pandemic, radiologists have used AI to determine COVID-19 diagnoses and progression using a variety of radiological evidence (12–14). One study found that using AI to interpret radiological data improved COVID-19 diagnosis when compared with a radiologist-only approach without the aid of AI (15). Another study found that AI can be a very useful application in terms of improving image interpretations in distinguishing lung viruses, such as COVID-19, from other pneumonia diseases with high accuracy (16). Radiographers are in charge of image post-processing, such as 3D image formation or multiplanar reconstruction, which is mostly automated and can be improved with AI integration (17). Some radiographers may have exciting or scary views of AI, and these could be heightened by the thought of having an “AI colleague” in the radiology department (18). The integration of AI into radiology practice as a strategic plan will increase the possibility of its successful implementation. For AI applications to be well-integrated in the clinical radiology department, radiographers must support the integration process because they are the interface between their patients and the technology. However, limited studies exist involving radiographers and AI systems. Few studies have been conducted to better understand radiographers’ perspectives and readiness for the use of AI in medical imaging practice. These studies explored radiographers’ perspectives on the integration of AI into medical imaging practice to identify factors that could help improve the implementation process in Africa and Ghana (19, 20). Recently, two studies on this topic were conducted in Saudi Arabia (21, 22). These studies included students who lack experience in the real-life practice of a radiology department and might have affected the results. However, this study explores only radiologists’ and radiographers’ perspectives on AI in medical imaging practice in Saudi Arabia, because they are at the forefront of this technological leap with real-life experience and adequate practice. Understanding their views, in particular, is key for the optimal development and implementation of AI in medical imaging. Therefore, this study aimed to explore the perspectives of radiologists and radiographers practicing in Saudi Arabia on the integration of AI into radiology practice in order to support policy development and enhance the AI implementation strategy for Saudi Arabia. OBJECTIVE This study aimed to explore the perspectives of radiologists and radiographers practicing in Saudi Arabia on the integration of AI into radiology practice in order to support policy development and enhance the AI implementation strategy for Saudi Arabia. METHODS 2.1 Study design, sample size and ethical considerations The study was designed as a self-administered survey that was distributed electronically between March 2021 and November 2021 to reveal the perspectives of radiographers and radiologists on AI in Saudi Arabia’s radiology departments. The required minimum sample size (n = 150) for this study was calculated using the G*Power version 3. In terms of the data collection procedure, Google Forms was used to present the survey electronically. Participants were mainly reached via social media, including WhatsApp, Facebook and Twitter. The first page of the survey contained an introductory information sheet that explained the purpose, risk, benefit, study duration, what AI was about for radiographers and radiologists and the opportunity to withdraw from the study at any time. Furthermore, the first page of the survey required each radiographer to electronically provide consent for their participation for access to the survey. The survey population did not include any at-risk groups, and all participants were assured anonymity, so an ethical committee review was not deemed necessary. 2.2 Instrument The cross-sectional survey used in the current study was previously validated and tested by a panel of academics with 7–10 years’ experience in radiography research and practice (19). The survey consists of several sections: (a) questions regarding demographic data, (b) attitudinal perspective items (five Likert scale statements) towards AI in medical imaging, (c) perspective items (ten Likert scale statements) on the positive impact of AI on medical imaging, (d) perspective items (eight Likert scale statements) on the negative impact of AI on medical imaging, (e) perspective items on factors affecting AI (four Likert scale statements) and decision making of AI (three Likert scale statements) in medical imaging and (f) one open-ended question (free-text comment) at the end of the questionnaire. The questionnaire had 36 items, including closed-ended questions and 5-point Likert scale statements (1 = strongly disagree to 5 = strongly agree). 2.3 Data analysis In the current study, the Statistical Package for the Social Sciences (SPSS version 27) was used for data analyses (for both descriptive and inferential analyses). Descriptive statistics were performed to generate frequencies, percentages and means, while inferential statistics were used to generate correlation coefficients and P-values. The responses to the rating items were assigned scores (1–5) on the Likert scale, corresponding to responses (strongly agree = 5, agree = 4, neutral = 3, disagree = 2 and strongly disagree = 1). For easy presentation of the results in tables, the strongly agree and agree responses were grouped together (coded in one number). Similarly, responses for strongly disagree and disagree were also grouped together. The aggregate mean scores were generated for the study themes or components (scores accumulated by one factor). Spearman’s correlation was used to assess the relationship between radiographers’ perspectives on AI and their demographic characteristics. The Mann-Whitney U test was used to independently test the perspective variables against gender, age and job (employment) title categories since the data variables were non-parametric. A p-value of less than 0.05 was considered statistically significant. RESULTS 3.1 Demographics A total of 173 respondents (74.1% males, 25.3% females, 77% radiographers, and 23% radiologists, with a mean age of 31.4 ± 5.3 years) working in radiology departments in Saudi Arabia responded to the current cross-sectional survey. Table 1 shows a summary of the respondents’ demographic characteristics. 3.2 Respondents’ perspectives on the clinical application of AI, the positive and negative impacts of AI, and factors that can affect AI implementation in medical imaging The findings of the respondents’ attitudinal perspectives on the clinical application of AI in medical imaging are described in Table 2. The respondents who were aware of AI as an emerging trend in medical imaging, excited about the emergence of AI in medical imaging and embraced AI technology as the future in medical imaging account for more than 56%. In sum, the respondents scored AI on an average of 1.7 on a scale of 1–3, suggesting a positive attitude towards the integration of AI technology in medical imaging. The respondents’ perspectives on the positive and negative impacts of AI in medical imaging are described in Tables 3 and 4. More than half of the respondents (55.5%) indicated that AI would have an overall positive impact on medical imaging practice. The respondents indicated that AI could be an assistive tool to ease radiographers’ work, increase access to patient’s care and improve decision-making, quality assurance, research productivity and accuracy levels in diagnosing diseases and education (Table 3). Furthermore, 63% of the respondents indicated that AI would reduce radiation dose levels while maintaining optimal image quality. On the other hand, the majority of respondents (nearly 60%) expressed fears about the potential machine errors associated with the use of AI-integrated equipment in radiography practice (n = 102) (Table 4). On a scale of 1–3, the respondents scored the positive and negative impacts of AI at an average of 1.6 and 1.8, respectively. The findings of respondents’ perspectives in relation to factors that can affect AI implementation and decision-making with AI in medical imaging are described in Tables 5 and 6. A lack of knowledge (43.9%) and perceived cyber threats (37.7%) were some of the factors identified as hindering AI implementation in Saudi Arabia (Table 5). In terms of decision-making with AI in medical imaging, nearly half of the respondents indicated that diagnostic decision-making should remain a human task and should not be handled through an AI algorithm (Table 6). 3 Diagnostic decision making should be handled by AI algorithms. 52 (30.1%) 53 (30.6%) 68 (39.3%) 3.3 Associations between respondents’ perspectives and demographic characteristics There was a significant negative association between respondents’ perspectives on the positive impact of AI implementation and education or qualification level (p = 0.02). In addition, there were significant positive associations between respondents’ perspectives on the negative impact of AI implementation and age and job title (radiographer or radiologist) (p = 0.04) for all. There were no significant associations between any respondents’ perspectives and the type of hospital where they worked (governmental, private or military), type of modality and previous experience in computer coding (p > 0.05). 3.4 Differences in respondents’ perspectives related to gender, age and employment title There was a statistically significant difference between radiographers and radiologists in terms of their perspectives on the negative impact of AI (p = 0.048). There were no significant differences between males and females or those aged above or below 40 years in terms of their attitudinal perspectives towards AI, perspectives on the positive impact of AI implementation and perspectives on the factors affecting the implementation or diagnostic decision-making of AI (p = > 0.05). 3.5 Open question (free-text comments) The open-ended question resulted in some themed free-text comments provided by the respondents relating to AI in medical imaging practice. Some of the respondents supported the implementation of AI in medical imaging, whereas others commented on a lack of knowledge about AI and the need for further training. One of the respondents commented on the need for AI in mammography. CONCLUSIONS This survey was developed to begin gauging the radiographers’ and radiologists’ perceptions, level of understanding, concerns and opinions on the emerging use of AI in medical imaging practice, research and training. Despite the fact that AI has only recently been introduced in the field of diagnostic radiology (23), this study revealed that the majority of participants had at least basic computer programming or coding experience (81%). More than half of the participants (57%) expressed an awareness and excitement about the emerging trend in the field of medical imaging, with 56.6% embracing it as the future of the discipline. This is consistent with the views reported in previous studies (19, 24). In general, the respondents scored AI a mean of 1.7 on a scale of 1–3 to suggest a very positive attitude towards AI in radiology. However, no apparent statistically significant association between respondents’ attitudinal perspectives and their demographic parameters, such as their level of education (P = 0.15) and years of work experience (P = 0.65), was observed. Understanding the implications of AI is critical for medical practitioners, particularly the technology’s meaning and contribution to the radiology profession. According to experts, AI-based applications will alter the ethical, scientific, economic and clinical future of radiology (25). In terms of the positive impact of AI, more than half of the participants reported that AI could be an assistive tool to ease their work, optimise radiation dose levels (63%), improve quality assurance (62%), increase research productivity and, in general, have an overall positive impact on medical imaging, which is consistent with several other previous studies (10, 18, 19). Not only does AI have a positive impact on clinical practice from the respondents’ perspectives, but also on the academic field. AI tools are thought to improve medical imaging education and promote radiology research productivity in the academic field, which is in line with Sarwar et al.’s findings (24). Meanwhile, there were no apparent statistically significant associations between respondents’ perspectives on the positive impacts of AI and their demographic parameters (P-values > 0.05) In terms of AI’s negative impact, respondents scored the technology a mean of 1.86 on a negative impact scale ranging from 1–3, indicating that they do not have concerns about AI. For example, 48.6% of respondents did not believe that AI would eventually displace them in a clinical environment. This was consistent with the findings of a survey conducted among medical students, who largely refuted the perception that radiologists would be replaced in the future (26). This may be because a majority of the respondents had a basic knowledge of computer coding. Furthermore, more than half of the respondents were aware of this technology. These findings contrast with a previous study (19). However, a majority of respondents (n = 102) were concerned about the possibility of machine errors associated with AI-induced equipment in the radiography unit and 42.8% believed that AI might change their role in the unit. This is in line with other studies’ findings (19, 24, 27). An explanation for this belief could be a lack of sufficient and in-depth understanding of how AI is implemented and what it can achieve beyond implementation. Furthermore, there was no statistically significant association between respondents’ perspectives on AI’s negative influence and their demographic characteristics (P > 0.05), except for age and job title (p = 0.04), respectively, which implies that all respondents would need similar training, irrespective of gender, to alleviate some of their negative perspectives on AI. With respect to the factors that can affect the implementation of AI in medical imaging, the majority of the participants acknowledged that the lack of robust cyber security measures (63%), compared with the other participants’ opinions and knowledge on the emergence of AI technology, poses a significant barrier to AI implementation in Saudi Arabia (44%). The results of this study are in line with several other studies (19, 28). Only 31.2% of the participants stated that the high cost of AI systems would impact its implementation in Saudi Arabia. A previous study in Ghana reported that 78.1% of the respondents stated that high equipment cost was a major factor to hindering AI implementation (19). There was no association between the respondents’ perspectives on factors affecting the implementation of AI and demographic characteristics. As to who should make the decisions in the use of AI tools, nearly half of the participants (n = 72) believed that radiology errors made in cases with AI-platform contributions, respondents and vendors should be held equally liable, and a small minority even believed that the vendor alone should be handled by the AI-algorithm. The remaining half stated that the responsibility for diagnostic decision making should remain a human task, which is consistent with the findings from Sarwar et al.’s study (28). This could be because 60% of the respondents’ stated that AI systems are merely supporting tools. There are several limitations to this study. The sample size was small, and the number of study participants who used AI in their clinical practice was not reported. Conclusion In this study, radiographers and radiologists in Saudi Arabia demonstrated a positive attitude towards the integration of AI into medical imaging; however, concerns regarding AI-related errors, cyber security, data protection and decision-making issues were identified. Understanding radiologists’ and radiographers’ perspectives on artificial intelligence in medical imaging in Saudi Arabia has significant implications for practice, ensuring optimal technology development, implementation and training, as well as planning for prospective role changes. |
