I believe artificial intelligence, more specifically deep convolutional neural networks (dCNN), could be used to classify US breast lesions. In our study, the implemented dCNN, with its robust sliding window approach, has demonstrated that high accuracies in the classification of US breast lesions can be reached using AI.
In order to be integrated into the clinical workflow, classification strategies based on artificial intelligence and convolutional neural networks, In my opinion, should rebuild and mimic the human decision process, rather than trying to provide a tool for the binary decision between malignant and benign lesions, which is futile in most circumstances.
This implementation into the clinical workflow may lead to a reduction of false positive lesions and entail a more accurate estimation for short-term follow-up examinations while simultaneously reducing morbidities and overall healthcare costs. In conclusion, I think artificial intelligence cannot substitute clinicians, as the diagnosis by radiologists do not only rely on the evaluation of image data, but also take into account patient familiar history, previous examinations and complementary imaging data as well as age or comorbidities. However, I am convinced that in the future, radiologists not using artificial intelligence in their clinical routine will sooner or later have to adapt to this rapidly developing technology and integrate AI into their daily workflow.
- Deep convolutional neural networks could be used to classify US breast lesions.
- The implemented dCNN with its sliding window approach reaches high accuracies in the classification of US breast lesions.
- Deep convolutional neural networks may serve for standardization in US BI-RADS classification.
Authors: Alexander Ciritsis, Cristina Rossi, Matthias Eberhard, Magda Marcon, Anton S. Becker, Andreas Boss