On October 5th, the Radiological Protection Study Group of the São Paulo Society of Radiology (SPR) was held with the theme “Artificial Intelligence Applied to Radiodiagnosis”. The presentation was conducted by Ana Maria Marques, PhD, nuclear physicist from the University of São Paulo (USP) and professor at PUC in Rio Grande do Sul.

Ana Maria, PhD, began with an overview of how Artificial Intelligence was present in radiologists’ discussions, especially after the 2018 RSNA, which focused heavily on this topic. She took the opportunity to remember the phrase of Curtis Langlotz, MD, who says “the AI will not replace radiologists, but the radiologists who use it will replace those who do not use it”.

To demonstrate the importance of artificial intelligence in current procedures, she brought a study that took place in the United Kingdom and the United States. More than 28,000 women underwent mammography and the professionals analyzed the ability of the AI to identify the presence of breast cancer in tests proven by biopsy or normal imaging.

It has been found a perform similarly or better than experienced radiologists. Below, the evolution of artificial intelligence performance over the years and what we can expect for the future:

As AI depends on a database to function correctly, the increased acquisition of medical data, such as images, signals, laboratory tests, reports and electronic medical records, and the increase of computational capacity (storage, communication and processing) result in its growth. It can be concluded that the development of the medical image processing and analysis area was the main point for the evolution of AI.

Artificial Intelligence as an ally of health professionals

With the clarity that problems can be solved through computational powers, artificial intelligence begins to be more talked about, mainly due to the market’s performance. With the boom kicked off in 2017, many startups emerge in the healthcare area and offer the assistance software for image analysis.

Artificial intelligence is a huge area and from it come smaller sub-areas, such as Machine Learning (ML), in which systems must learn over time by extracting features and selecting the most important variables to create a model of classification.

Within ML, Deep Learning (DL) appears, in which the system tests many features that are not related to human knowledge and creates a model, without the need for professionals to extract the features. The approach should be chosen according to the problem at hand.

This set of systems that help in medical diagnosis is important because it unites man and machine and creates an integration capable of reducing false positives.

One type of research that helps in detecting irregularities is Radiomics, mentioned at the Study Group meeting, in which information is extracted and analyzed according to the universe of radiomics attributes.

Therefore, it’s not only the image that is interpreted, but also the values of shape, intensity and texture, which has numbers assigned. This look at another level is similar to genomics, which can see similarity between a group of patients according to the attributes discovered in the analysis. For this, data curation is important, because if they are bad, the result will not be good.

There are still some challenges for the full implementation of artificial intelligence in the doctors’ daily life, such as the lack of standardization in the stages of data and image acquisition, irregular quality, few studies that assess uncertainties due to segmentation and pre-processing techniques , little explanatory research, among others.

However, it is important for optimizing the diagnosis by extracting information that goes beyond visual interpretation, adding value with “invisible” quantitative and radiomic information.

Artificial intelligence should be an ally of professionals, not a competitor. For this, they must inform themselves and take ownership of the matter so that it can be implemented with balance and wisdom, protecting the main focus of this process: the patient.

Check out the full presentation, in Portuguese: