AI in Head and Neck Cancer - Why are we still waiting for the breakthrough?
The Application of AI in Head and Neck Cancer
My focus on AI is primarily in radiology, given my background, so I acknowledge a certain bias in my perception of how extensively AI is utilized in clinical settings (though "extensively" might be an overstatement considering the current minimal routine use of AI in radiology). This is why the survey conducted by Giannitto and his team from Italy particularly caught my attention. They distributed a survey within the Head and Neck (H&N) cancer community to gauge awareness and attitudes towards AI. This allowed them to identify non-technical challenges that AI implementation might face on the human side, particularly when considering broader usage of these tools. In summary, the survey highlighted a significant lack of knowledge and awareness about AI and its potential to improve clinical practice.
Use Case for AI in H&N Cancer
The majority of H&N cancers (two-thirds) are diagnosed at an advanced stage and require a multimodal approach by a multidisciplinary team. AI tools seem ideally suited to manage the vast amounts of data involved in the entire diagnostic workup and treatment process and to draw accurate conclusions from it. However, the implementation of AI in H&N oncology has been slow. So, why is that?
Study Design
Giannitto and his team disseminated surveys via social networks and email lists. These surveys consisted of 19 questions and gathered responses from 139 participants.
Who is Using AI and What is it Mainly Used For
Approximately 50% of all participants have used AI at some point, with 66% of these users being radiologists. Remarkably, 50% have never used any AI in their medical practice, and 8% (!) were not familiar with the term. Otolaryngologists accounted for 16%, while oncologists reported no AI use. Furthermore, the typical AI user is likely to work in an academic hospital (68%) and is either an attending or a consultant (50%), generally younger.
The predominant use of AI by radiologists suggests that imaging is the primary application for AI, which aids in pattern recognition and feature classification. Common clinical use cases mentioned included:
- Prediction of outcomes and complications
- Patient information and shared decision-making
Potential Use Cases Predicted by Survey Participants
Interestingly, the top potential use cases envisioned were in the field of surgery:
- Pre-planning
- Reduction of procedural times
- Assistance in skull base surgery
This stands in stark contrast to the current situation, where AI has yet to significantly impact the field of surgery.
Causes of Reservations Towards AI
42% of participants expressed reservations about using AI. The top concern was the fear of error and misdiagnosis (57%). Additionally, there was worry about the potential depletion of medical knowledge among professionals and concerns about losing control over AI-driven processes in diagnostics and treatment. Another significant concern was the potential disruption of the patient-doctor relationship by AI.
The authors concluded that, if used correctly, AI could be a game changer.
My Take on It
I fully agree with this conclusion. Each field in medicine needs to identify and define its specific use cases for AI. A thorough and continuous risk assessment must be part of this process. As AI becomes more accurate and data availability grows, perspectives will shift—soon, manually performing tasks without the support of readily available AI may be considered a risk.