Weaknesses of AI Surgical Training

The original use for AI in surgical education was to provide individualized feedback to students, and very little attention was paid to actual assessments of students learning. Since then many improvements have been made to AI including the development of a teaching assistant that can teach material as well as supervise students and provide feedback. However as mentioned in my previous blog post, although surgical simulations used as training for surgeons can be extremely beneficial, the virtual operating assistant along with other surgical training tools that use AI algorithms, do possess certain weaknesses which are explained in a research study performed by Kai Siang Chan and Nabil Zary. The inability to provide sufficient feedback, the possibility of “cheating the system”, the presence of board parameters, and the lack of surgeon expertise and involvement are the few of the obstacles that stand in the way of AI surgical training tools from being implemented in the real world.


One major challenge when using AI as a teaching instrument, is the feedback, or lack of feedback, that students receive from it. Feedback is one of the most important aspects of learning as it assists in identifying knowledge gaps and goals for learning so that students can take the steps necessary in order to improve their skills. However, while one of the main reasons for AI use is its ability to provide instant feedback, AI algorithms often lack detail and corrective criticism in their reasoning. In many cases, AI systems fail to provide an explanation of how a certain answer or prediction was reached. This issue, otherwise known as “explainability” is part of a subcategory of AI called deep learning (Chan & Zary, 2019). An explanation is crucial in the learning process, specifically in medical training, so AI systems have to be able to understand the emotional and cognitive thinking of learners and provide appropriate feedback.


Another important challenge that needs to be addressed is the ability of an AI-powered teaching platform to allow students to cheat the algorithm. It is not certain whether such algorithms actually train surgeons and increase their skill level or just provide a foundation for users to “cheat” the algorithm and give the impression that they are skilled in a certain area. Overcoming this challenge provides the perfect opportunity for humans instructors and experts to give their input and invest themselves in the learning process as well. If such platforms undergo meticulous training and validation with the help of expert options and other research studies that can properly assess users as if they were in a real life situation, then cheating the system would be much less likely. A feature of the Virtual Operative Assistant which makes it easy to cheat, are the relatively broad parameters that allow one to be classified as either skilled or novice. In the experiment completed during the Virtual Operative Assistant, there was a misclassification where 4 novice participants were actually labelled as skilled (Mirchi et al., 2020). In order to have a more efficient classification system, human experts can be used to determine and help the Virtual Operative Assistant recognize specific markers of a good surgeon as well as appropriately distinguish between novice and skilled surgeons.


Human expertise or interaction of any kind can prove to be extremely beneficial when working with AI and machine learning algorithms, especially in the surgical training field. On occasion, if the learner feels a disconnect from their teacher, or the feedback is not properly backed up with reason, then that feedback can have reverse effects and actually be unproductive or damaging to the students skill level (Chan & Zary, 2019). Furthermore, if there is an error in the coding of an AI system and the results are incorrect, students can be negatively impacted. AI-based platforms used for teaching should be integrated into a special educational system with a combination of machine learning and human interaction because although AI can provide automated feedback as well as straightforward and standardized training, human to human connection has great importance in education. If scientists would like for AI-based surgical training to be effective and implemented into the real world in order to cure diseases such as cancer, then the educational systems that they are incorporated into, would need to be thoroughly and precisely constructed and evaluated. For these reasons, AI could be substantially more useful for tasks where they are able to assist humans or when humans are not able to perform the task at all, such as computerized testing, programmatic assessment, cancer screening/diagnosis, and cancer treatment.


References

Chan, K., & Zary, N. (2019, June 15). Applications and Challenges of Implementing Artificial Intelligence in Medical Education: Integrative Review. Retrieved August 10, 2020, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6598417/

Mirchi, N., Bissonnette, V., Yilmaz, R., Ledwos, N., Winkler-Schwartz, A., & Del Maestro, R. (2020, February 27). The Virtual Operative Assistant: An explainable artificial intelligence tool for simulation-based training in surgery and medicine. Retrieved August 10, 2020, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7046231/

Comments

Post a Comment

Popular posts from this blog

The Impact of AI on Jobs

The many promises of AI included automated image segmentation, detection of cancerous lesions, and comparison of images. AI systems will not get tired while performing the same tasks that doctors may get tired and lose accuracy while completing. AI can also help physicians by completing tedious tasks and managing large amounts of data. Depending on the requirements of a job, a worker can either be helped by technology or in competition with it. For example, improvement in robotics can decrease opportunities for employment for workers in manufacturing but in the field of artificial intelligence and cognitive technology, it would increase the demand for workers. Recent data expresses that when considering its image and predictive analysis, AI might soon prove to be more efficient than radiologists. However, it is likely that AI will not replace the role of general physicians, but rather augment them. This is due to the fact that AI is unable to engage in interactions with patients that
Read more

Segmentation and Cervical Cancer Classification

Cervical cancer does not occur as often as brain and lung cancer, however it still possesses an incredible threat to women every year. While it can be cured if found in an earlier stage, many women die every year due to the fact that their cancer was not detected at an earlier time and symptoms did not appear until the cancer was too far advanced to treat. Cells in the cervix can either by squamous cells, which when infected cause squamous cell carcinoma, or glandular cells which cause adenocarcinoma. Squamous cell carcinomas are flat cells that vary in morphology and cover the surface of the cervix. Adenocarcinoma cells develop in the glandular cells which produce music and grow along the inside of the endocervix, the passage that runs from the cervix to the womb. While both squamous cells carcinomas and adenocarcinomas are both associated with HPV infection, adenocarcinomas are much less common (Conceição, Braga, Rosado, & Vasconcelos, 2019, p. 2) . Because cervical cancer is dif
Read more

Doctor-Patient Feedback and Interpretation

Of the different challenges that AI algorithms in medical training pose, feedback and interpretability are two of the most prominent issues. Interpretability of AI algorithms is the ability of a human to understand the way it made a connection between features extracted and its predictions. A study done at Mount Sinai Hospital created a type of AI algorithm known as deep learning using data from 700,000 patients. Their algorithm was highly accurate and was able to diagnose conditions that even experts struggle to diagnose, such as  schizophrenia. However, there was no way for humans to know how the system reached a diagnosis (Paranjape, Schinkel, Nannan Panday, Car, & Nanayakkara, 2019) . This is an issue because it becomes very hard for patients to trust a system that cannot provide an explanation and if a calculation were to be made incorrectly that puts a patient in danger, then it is not known whether the doctor, the hospital, or the company that developed the AI algorithm is l
Read more