TL;DR
This study surveyed European gastroenterologists specializing in capsule endoscopy (CE) to gauge their perceptions of Artificial Intelligence (AI). While generally optimistic about AI's potential to improve efficiency and accuracy in CE, most physicians do not believe AI will fully replace human readers in the foreseeable future.
What Problem Does This Paper Address?
Capsule endoscopy (CE) involves reviewing extensive video data, a time-consuming task for gastroenterologists. Artificial intelligence (AI) has shown promise in automating aspects of this process, but the adoption of new technologies can be met with skepticism or apprehension. This paper addresses the gap in understanding physician sentiment towards AI in CE, exploring their expectations, concerns, and perceived barriers to implementation. Without this understanding, the successful integration of AI into clinical practice is hindered.
Methods
The researchers designed and distributed an online, anonymized survey to European gastroenterologists, specifically targeting those involved in capsule endoscopy reading (CERs). The survey consisted of 32 questions covering general information, interest in new technologies, and specific perceptions of AI in medicine and CE. Responses were collected using a Likert scale, and statistical analysis was performed using Rstudio. The study also involved a qualitative component where national leaders of the International CApsule endoscopy REsearch (I-CARE) group helped disseminate the survey within their communities.
Datasets
Not discussed in the paper. The study relied on survey responses from 380 gastroenterologists across 20 European countries.
Key Insights and Ideas
- AI as an Enhancer, Not a Replacer: The overwhelming sentiment is that AI will be a valuable tool to assist, not replace, human CE readers. This suggests a focus on human-AI collaboration rather than full automation.
- Efficiency is Key: A major driver for AI adoption is the potential to significantly shorten CE reading times, a known bottleneck in current practice.
- Standardization and Characterization: Physicians see AI as a way to standardize reporting and improve the characterization of lesions, leading to more consistent and accurate diagnoses.
- Training is Crucial: There's a strong interest in both general and advanced AI training, indicating a desire to understand and effectively utilize this technology.
- Patient Impact is Considered: While most believe AI won't change the physician-patient relationship significantly, a notable portion anticipates a more interactive dynamic.
- Cost is a Real Concern: The financial implications of implementing AI solutions are viewed as a significant factor by a majority of respondents.
Key Results
- High Agreement on AI's Positive Impact: 80% of CERs agreed or strongly agreed that AI would positively impact CE.
- Shortened Reading Time Expected: 79% agreed or strongly agreed that AI would shorten CE reading time.
- Standardization and Characterization: 74% and 68% agreed or strongly agreed that AI would help standardize reporting and characterize lesions, respectively.
- No Full Replacement: 71% of CERs at least moderately disagreed with the idea that AI would replace them at work.
- Interest in Training: 75% were interested or strongly interested in generic/baseline AI training, and 68% in technically advanced training.
- Operator Dependence Concern: 60% agreed that AI implementation could lead to operator dependence.
- Cost Importance: 65% agreed or strongly agreed that the cost of AI implementation is important.
Limitations and Future Work
The authors acknowledge several limitations:
- Participation Bias: The survey was voluntary, meaning respondents might be more interested in AI, potentially skewing results towards optimism. Academic physicians were overrepresented.
- Online Survey Limitations: The lack of a physical meeting meant no real-time debate, which could have enriched the discussion.
- Definition of AI: The broad definition of AI used might have influenced responses.
- No Participation Rate: The method of survey distribution made it impossible to calculate a participation rate, potentially leading to non-response bias.
Future work could involve:
- Medico-economic Studies: Evaluating the cost-effectiveness of AI solutions in CE to determine feasibility and safety.
- Further Research on AI Performance: More studies are needed to demonstrate the performance of AI devices, especially with external validation.
- Developing AI Techno-vigilance: Creating systems to monitor and address potential harms from AI in medical applications.
- Mapping Research Gaps: Continuing efforts to identify research, clinical gaps, and challenges for AI in CE.
Summary updated