Treatment Candidates to Avoid Cardiotoxicity from Cancer Therapy by Explainable AI
This study aims to propose a potential treatment that can avoid the onset of cardiotoxicity derived from cancer treatment by utilizing the national hospital cancer registry data, the largest cancer treatment database in Japan, DPC data, and explainable artificial intelligence (XAI).
Cancer treatment has steadily been progressing. However, cardiotoxicity, side-effects of cancer treatment (especially chemotherapy) on the heart, has become an issue, which encourages us to study cardi-oncology. In addition, adaptation of AI in the medical field is being considered, but there is the problem of the 'black box' nature of AI models (how information is processed inside AI is difficult for humans to interpret). In the medical field, it is a big problem not to know the basis for treatment decisions and decision making. XAI is a method to reduce the black box nature of AI models and improve their interpretability. In particular, this research will focus on a method to derive a modification plan to change the prediction result of the AI model from "onset" to "non-onset" on an individual basis.
If this research is successful, it will generate a hypothesis for a treatment method to avoid the onset of cardiotoxicity due to side effects of cancer treatment, and testing the hypothesis can be used for future research. In addition, if the XAI of the present study can overcome the black box nature (difficulty in interpretation), we believe that it will promote the use of AI in other areas of medical research.
If the present study succeeds, we can hypothesize a treatment that avoids the onset of cardiotoxicity due to side-effects of cancer treatment. This may lead to related research and the development of therapies to avoid cardiac side-effects of cancer treatment. If the XAI used in this research can overcome the black box nature of AI, the application of AI in the medical field and clinical practice will be advanced. We will be able to support the decision-making process of doctors and patients by calculating diagnosis and treatment methods tailored to individual patients based on large amounts of data and presenting them to doctors.
Comments from principal researcher
I would like to help people undergoing cancer treatment from the point of view of cardiology and data science. Thus, I have started the present study combining cardiology, oncology, and data science. It must be important to check whether XAI can work well in the medical field. Through this research, we hope to reduce the side effects of cancer treatment on the heart and promote the use of AI in the medical field.
In the slide, section 1, the text under the blue title, left-hand side: the text here looks bad as the word-breaks to fit in the available space are incorrect, i.e. Backgrou-nd, instead of Back-ground; Treatmen-t, instead of treat-ment