ARSER’s main objective is to develop research into the after-effects of radiotherapy, and thus, to achieve this, to raise funds.
Its aim is not to establish basis for claim: ARSER is not intended to group patients considering they are “victims” of radiotherapy. Rather ARSER wants to inform patients and doctors about the progress of the research on the after-effects of radiotherapy and to finance it.
Dr. Sylvie Delanian is one of the very first physicians in the world to have cared about the after-effects of radiotherapy as a radiotherapist and to have provided answers, particularly for treatments that have been in effect for more than 10 years.
Along with Jean-Louis Lefaix, a researcher at the CEA, and then with Dr. Pierre-François Pradat, neurologist, she led research to better understand the mechanisms and develop new treatments to reduce these after-effects.
Many patients cured of their cancer long ago, called “long-term survivors”, and no longer undergoing medical treatment, are often unaware that the ill-systematised symptoms they suffer are after-effects of radiotherapy.
by Jean-Pierre Briot
Emeritus Research Director, CNRS – Sorbonne University
Member of the ARSER Board
The purpose of this editorial is to take stock of the advances and prospects for the use of artificial intelligence techniques for medicine, and in particular the after-effects of radiotherapy.
Allow me first of all to recall that AI (artificial intelligence) is one of the fields of computer science and was projected from its inception, in the 1950s. It includes (but is not however reduced to ) statistical learning techniques (known as “artificial neural networks”), which have experienced dazzling and constant progress since 2012, due to the conjunction of three factors: technical progress, availability of huge masses of data thanks to to the Web, and availability of large resources for intensive calculations thanks to cheap adapted processors. Examples of everyday applications are face recognition, translation and weather forecasting.
Rather than write this analysis by myself, it seemed tempting to ask ChatGPT, the conversational agent created by the company Open AI, and very publicized for a few months. Before that, it seemed useful to me to summarize the principles of its operation.
– Its first GPT module (acronym for Generative Pre-trained Transformer) has the function of statistical language learning (being able to dialogue in several languages, including French). GPT is a very large (so-called deep) artificial neural network architecture. Its training is of the “supervised learning” type, consisting in learning to predict the following words of a sentence, from an enormous quantity of texts, and thus the syntax and the structure of the language. The result is an encyclopedic text generator.
– The second module has the function of dialogue generation and uses another machine learning technique, called reinforcement learning , where human experts evaluate the responses generated by the system, allowing it to gradually learn responses and dialogues more and more refined.
The results are quite impressive, even if limits, some deep, exist and could be the subject of another editorial! Good reading !
Below is the full transcript of the dialogue initiated with ChatGPT on March 23, 2023
– Question (Q): Please summarize the main advances from AI to medical progress
– Answer (R-IA): AI has brought many advances in the field of medicine, including:
- Computer-aided diagnosis : AI systems can help diagnose diseases by analyzing medical images such as X-rays, CT scans and MRIs, as well as the patient’s symptoms and medical history.
- Early disease detection : AI algorithms can be used to detect early signs of diseases such as cancer and cardiovascular disease, enabling more effective treatment.
- Personalization of treatments : AI can be used to analyze patient data and provide personalized treatment recommendations, based on medical history, symptoms and test results.
- Medical robotics : Robots can be used in surgery, helping surgeons perform more precise and safer procedures.
- Patient monitoring : AI can be used to monitor patients remotely, detect changes in their health status and prevent complications.
- Drug discovery : AI algorithms can be used to identify promising compounds for drug development, by analyzing large amounts of data.
These AI advances in medicine have the potential to transform the way diseases are diagnosed, treated and managed, enabling more precise and personalized medicine.
– (Q): Thank you, do you have more specific information on how AI could help cure the side effects of radiotherapy when it has been used to fight cancer?
-(R-IA): Yes, AI can help cure the collateral effects of radiation therapy by optimizing the radiation dose delivered to surrounding healthy tissue during cancer treatment.
Irradiation of healthy tissues can cause side effects, such as skin burns, pain and organ function disorders. Doctors must therefore strike a delicate balance between the dose of radiation needed to destroy cancer cells and the dose that can damage surrounding healthy tissue.
This is where AI can come in. AI systems can analyze patient medical images, such as CT scans and MRIs, to map surrounding healthy tissue and determine areas that require radiation dose reduction. Then they can use optimization algorithms to adjust the dose of radiation delivered to healthy tissue while maintaining a sufficient dose to treat cancer.
By using AI to optimize radiation dose, doctors can reduce side effects for patients, improve treatment outcomes and reduce healthcare costs. Ultimately, it may help improve the quality of life for cancer patients and improve survival.
– (Q): More specifically, we could classify the types of collateral problems of radiotherapy into two types: (1) fibrosis, on the skin, nerves and blood vessels and (2) necrosis, on the bones. How do you think AI could help cure or at least mitigate each of these two types of problems?
– (R-IA): (1) Fibrosis : AI can help to cure or mitigate the collateral effects of radiation therapy on fibrosis by optimizing the radiation dose delivered to surrounding healthy tissue. AI systems can analyze medical images to map surrounding healthy tissue and determine areas that require radiation dose reduction. Optimization algorithms can then adjust the radiation dose delivered to healthy tissue while maintaining a sufficient dose to treat cancer. By reducing the radiation dose to healthy tissue, AI can help reduce the likelihood of developing fibrosis, which can help improve patients’ quality of life.
(2) Necrosis: Bone necrosis is a rare but potentially serious complication of radiation therapy, which can cause severe bone pain and fractures. AI can help identify patients who are most at risk of developing bone necrosis after radiation therapy, by analyzing medical images to assess bone mineral density and other risk factors. This allows doctors to monitor patients closely and take preventative measures, such as using drug treatments to strengthen bones. Additionally, AI can be used to optimize the radiation dose delivered to bone, reducing the risk of developing bone necrosis while maintaining a dose sufficient to treat cancer.
– (Q): Thank you very much for this information. I am going to show your text as it is for an editorial on the website of ARSER, our research association, which stands for Association for Research on the Sequelae of Radiotherapy, mentioning of course that you are at the origin of these informations. You are thus demonstrating a potential advancement of AI to medical progress (and any field) in the interactive and personalized delivery of information.
– (R-IA): I am delighted to have been able to help you with this information. Feel free to use this text for your editorial on ARSER, mentioning of course that the information was provided by ChatGPT, an artificial intelligence trained by OpenAI. I am convinced that AI can play an important role in improving healthcare by providing advanced analysis and treatment tools to help cure the side effects of radiation therapy and other medical treatments.