Impact of artificial intelligence on radiology: a EuroAIM survey among members of the European Society of Radiology

Abstract

We report the results of a survey conducted among ESR members in November and December 2018, asking for expectations about artificial intelligence (AI) in 5-10 years. Of 24,000 ESR members contacted, 675 (2.8%) completed the survey, 454 males (67%), 555 (82%) working at academic/public hospitals. AI impact was mostly expected (≥ 30% of responders) on breast, oncologic, thoracic, and neuro imaging, mainly involving mammography, computed tomography, and magnetic resonance. Responders foresee AI impact on: job opportunities (375/675, 56%), 218/375 (58%) expecting increase, 157/375 (42%) reduction; reporting workload (504/675, 75%), 256/504 (51%) expecting reduction, 248/504 (49%) increase; radiologist's profile, becoming more clinical (364/675, 54%) and more subspecialised (283/675, 42%). For 374/675 responders (55%) AI-only reports would be not accepted by patients, for 79/675 (12%) accepted, for 222/675 (33%) it is too early to answer. For 275/675 responders (41%) AI will make the radiologist-patient relation more interactive, for 140/675 (21%) more impersonal, for 259/675 (38%) unchanged. If AI allows time saving, radiologists should interact more with clinicians (437/675, 65%) and/or patients (322/675, 48%). For all responders, involvement in AI-projects is welcome, with different roles: supervision (434/675, 64%), task definition (359/675, 53%), image labelling (197/675, 29%). Of 675 responders, 321 (48%) do not currently use AI, 138 (20%) use AI, 205 (30%) are planning to do it. According to 277/675 responders (41%), radiologists will take responsibility for AI outcome, while 277/675 (41%) suggest shared responsibility with other professionals. To summarise, responders showed a general favourable attitude towards AI.

Keywords: Artificial Intelligence; Machine Learning; Radiologists; Radiology; Surveys and Questionnaires.

Fig. 1

Distribution of survey responders according to age and sex

Fig. 2

Geographic distribution of survey responders

Fig. 3

Distribution of responders. The grey bars represent the number of responders that practice each subspecialty while the green bars represent those who foresaw an impact of AI on each subspecialty. Subspecialties are sorted according to the difference between values of green and grey bars

Fig. 4

Distribution of responders. Grey bars represent the number of responders that practiced each imaging modality, while the orange bars represent those who believe that that modality will be used to develop AI applications. Imaging modalities are sorted according to the difference between values of orange and grey bars. PET: positron emission tomography; DXA: dual X-ray absorptiometry; CT: computed tomography; MRI: magnetic resonance imaging

Fig. 5

Distribution of answers related to who should take legal responsibility of AI systems outcome