Hello BRIGHT Run Family,
Hope you are looking forward to having a nice April.
I would like to share some research findings with you.
This research deals with the determination of cosmesis, the aesthetic appearance of the breasts after breast-conserving surgery (BCS) followed by radiation therapy (RT). As you know, adverse cosmesis after BCS and RT can affect the quality-of-life, especially in younger patients.
In radiation-related clinical trials, a panel of physicians evaluates the post-treatment appearance of the breast using digital photos of breast cancer patients after their treatment.
Because humans are involved, this process is subjective, so more than one physician is engaged in the process to reduce the subjectivity. However, this also takes more time. We tried to resolve this by using artificial intelligence (AI) to determine cosmesis objectively.
To do this, we developed an AI tool with the help of deep neural networks, a type of AI model that learns from data. The model requires the image of a patient’s breast region, and it is guided to compare the differences between the treated and untreated breasts.
This advanced AI model can extract patterns from the images of treated and untreated breast regions, which may include variations in colour, shape, texture, and many more factors. Based on these variations, another AI model is used to categorize the cosmesis as adverse or non-adverse.
As I have expressed to you in my previous letters, we need data to develop the AI models. In this case, two radiation-based clinical trials from the Ontario Clinical Oncology Group (OCOG), which is affiliated with the Department of Oncology, Faculty of Health Sciences, McMaster University, provided the datasets. These two trials are called RAPID and OPAR. They contributed as the developmental datasets. Another developmental dataset was contributed by researchers at Portugal’s University of Porto. We were able to validate the performance of this tool with the help of images released from three other clinical trials.
Our methodology and results are published here: Measurement of adverse cosmesis in breast cancer: A deep learning approach – ScienceDirect. Our methodology is a step ahead from the methods in literature that require important key-points (such as detailed breast outlines) to perform cosmesis evaluation; just an approximate location of the breast region is sufficient for our tool to perform the evaluation.
I was supported in this study by my colleagues Dr. Whelan, Dr. Levine, Dr. Parvez along with Dr. Kong, radiation oncology fellow in the Department of Oncology at McMaster. Also, Mr. Philipchenko, IT Manager at OCOG, helped us to access the data.
While this tool is a step forward and demonstrates potential, we will need to run more experiments and studies before it could be used to help clinicians. I am happy to take a step in that direction through this work.
With brighter thoughts,
Best,
Ashirbani
Dr. Ashirbani Saha is the first holder of the BRIGHT Run Breast Cancer Learning Health System Chair, a permanent research position established by the BRIGHT Run in partnership with McMaster University.
