April 2021 saw the publication of journal articles that explored the application of artificial intelligence (AI) to breast ultrasound images. Writing in the Journal of Breast Imaging, Manisha Bahl of the USA asked, “Artificial intelligence for breast ultrasound: will it impact radiologists’ accuracy?” while Kazunori Kubota of Japan wrote an editorial in the Journal of Medical Ultrasonics entitled, “Breast ultrasound in the age of advanced technology and artificial intelligence.” A group from China led by Chuliang Wei published a paper in Computerized Medical Imaging and Graphics on tumour classification in automated breast ultrasound (ABUS) based on AI.
Kubota said that although hand-held ultrasound (HHUS) is one of the most commonly used modalities for detecting breast cancer, it has two major drawbacks: it is operator dependent and has low reproducibility. These problems have meant that ultrasound has not seen widespread use in mass screening programmes. Kubota believes that the introduction of ABUS, in combination with AI, will have a major impact on the way breast cancer is diagnosed and treated in the years to come.
Utilising an ABUS system manufactured by SIUI (seen left), the Chinese team studied 214 patients – 86 with malignant tumours and 128 with benign lesions. Once these imaging datasets had been carefully characterised by expert clinicians, convolutional neural networks were trained on a subset of 172 patients. Their AI algorithm was then tested on the other 42 patients and demonstrated 94% accuracy in classifying the lesions as either benign or malignant (see below right, © Elsevier).
In a recent commentary in The Lancet, Kerstin Vokinger from Switzerland explored the challenges faced by the Food and Drug Administration (FDA) in the USA as it considers the application of AI to medical devices. In her introduction, she stated, “Artificial intelligence (AI) and machine learning (ML) software have the potential to improve patient care. An underlying algorithm can either be locked so that its function does not change, or adaptive, in which the AI and ML system performs continual learning.”
Despite the obvious benefits of continual learning – also known as lifelong learning – the FDA has not yet approved medical devices based on this approach, although it has recently published an action plan. Continual learning does pose some risks, including errors in the new data and system deterioration if the new data are biased. A third risk is “catastrophic forgetting” in which new information could interfere with what the diagnostic model has already learned. We are nevertheless convinced that AI applied to ABUS images will lead to early detection of breast cancer in many more women.