We hear more and more often about artificial intelligence to improve diagnostics, and the breast field is no exception. But if for mammographic images artificial intelligence software and deep learning have already proven to effectively support human operators (increasing the sensitivity of the exam, saving time and optimizing resources), in the case of sine we are still in a running-in phase. And precisely to test the potential of one of these systems, to reduce overdiagnosis and, consequently, the number of patients referred for biopsy, the IRCCS Foundation of the National Cancer Institute in Milan has set up a study involving 240 patients. The results bode well for a future application of similar technologies in everyday clinical practice.
Breast cancer, with Artificial Intelligence the sensitivity of mammography increases by 11%
by Irma D’Aria
Breast ultrasound, too many false positives
Breast ultrasound is one of the tests that allow you to see the breast tissue, identifying the possible presence of abnormalities that require investigation. This methodology, however, is greatly affected by the human factor, the operator’s skill in carrying out the examination and his experience in interpreting ultrasound images. “Also for these reasons, breast ultrasound reports reflect an excess of false positives, i.e. suspicious anomalies that are directed towards the biopsy but which are not pathological. We record a rate double that of mammograms”, explains a Salute Gianfranco Scaperrotta, head of breast diagnostics at the National Cancer Institute of Milan. This brings with it a heavy load, both in terms of use of resources and anxieties and fears for the patients. “What we clinicians would like today is for artificial intelligence to help us break down this figure”.
From prevention to therapy, artificial intelligence revolutionizes the fight against cancer
by Valentina Guglielmo
AI for fewer “useless” biopsies
The premises to hit the target seem to exist. It was at the INT that a latest generation ultrasound system equipped with a sophisticated artificial intelligence system, branded GE Healthcare, was tested for the first time in Italy. The software is able to compare the ultrasound images submitted to it with approximately one million images in its database and to provide the human operator with a score on the scale of suspicion of nodule malignancy. In other words, it is as if the operator were asking the machine for a second opinion. “To verify the software’s performance, we designed a study protocol that involved 240 patients who had already planned a follow-up biopsy – explains Scaperrotta -. We submitted the ultrasound images of the patients to the machine and then compared the opinion of the artificial intelligence with the outcome of the biopsies”.
Breast cancer, so artificial intelligence will improve diagnoses
by Tiziana Moriconi
Preliminary data indicate an overall agreement of 85%, in line with performance obtained from other US and European experiences. “All in all, the machine performed well: false positives dominated the 15% of discrepancies, while false negatives accounted for only 5%”. On these bases it is estimated that by using AI to support the doctor, the biopsy rate could be reduced by up to 40%.
Pros and cons of AI today
Speaking of artificial intelligence, one always wonders whether machines will replace human beings, but, as far as ultrasound examinations are concerned, the clinician’s interpretation is still essential. “The most widespread and available artificial intelligence software today are ‘trained’ on thousands of archive images, but do not take into account many other factors which instead help to estimate the danger of a lump in the breast tissue”, explains the expert . “Information such as the patient’s age or a family history of breast cancer allow us to better evaluate individual situations, something that the software we have been able to test cannot do”. The experience of the clinician, therefore, is still a determining factor. “This is also demonstrated by the fact that, as we have noticed in our tests, artificial intelligence is more efficient when compared with the work of young doctors”, adds Scaperrotta. “Even within their current limits, therefore, machines can help us to render a more equitable service where there are shortcomings in the operators – whether due to lack of experience or typically human variables such as tiredness, which software is not affected by”.
Lung cancer, artificial intelligence can guide radiation therapy
by Valentina Guglielmo
Not all technologies that are developed and tested in the medical field turn out to be really useful in clinical practice. “However – concludes Gianfranco Scaperrotta – I believe we are at a turning point and that in the near future artificial intelligence will help us extrapolate more and more information from the images, for better diagnostics in every field”.