Comprehensive Summary
This study focuses on using AI to discover reliable non-invasive alternatives to the current methods of classification of Human epidermal growth factor (HER2), including invasive biopsy and immunohistochemistry, which are both subject to sampling bias and variability. An artificial intelligence framework was used to predict HER expression categories from dynamic contrast-enhanced magnetic resonance imaging. 3000 female breast cancer patients were included in the dataset, and an SSL framework was used to observe several million break DCE-MRI slices. The data shows that the model demonstrates robust performance across the used data sets. Furthermore, it was able to distinguish between different HER2 categories, especially between HER2-zero and HER2-low cases. It was discussed that of the patients predicted to be non-zero HER2, many were truly low or positive, indicating the ability of the model to identify patients who could potentially benefit from specific treatments. The study was able to enable more precise treatment strategies for breast cancer patients.
Outcomes and Implications
This research is relevant to the medical field as it is able to utilize a large data set to give an accurate model that can prevent patients from undergoing ineffective treatment. This will serve to benefit the population as more patients can receive the treatment they need and will work, as well as cut down on unnecessary treatments and resources. Furthermore, this advances the medical field as it demonstrates the ability to categorize different biological factors and potentially be used for other factors for different cancers.