AI screening of pap smears can detect breast cancer predators

Using artificial intelligence and a portable digital microscope, researchers hope to create screening devices that detect breast cancer predators in women in resource-limited settings. A study led by researchers at Karolinska Institutet in Sweden now shows that AI screening of pap smears performed with portable scanners was comparable to analyzes performed by pathologists. The results are published in the journal Open JAMA Network.

Our approach allows us to detect and treat predators to breast cancer, especially in low – income countries, where there is a high level of skilled pathology and advanced laboratory equipment. “

Johan Lundin, Corresponding Author, Professor, Department of Global Public Health, Karolinska Institutet

In countries with national screening programs designed to detect human cell and papillomavirus (HPV) anemia in cervical samples, the number of cases of cervical cancer has dropped dramatically. Nevertheless, the total number of cases is expected to increase over the next decade, largely due to a shortage of screening resources and HPV vaccines in low-income countries.

Innovative diagnostic solutions that take into account local conditions and limitations are needed if genetic screening is to be offered to more women worldwide.

For this study, the researchers trained on an AI system to recognize cell dysfunction in the cervix, which can be successfully treated when detected early. Smears were taken from 740 women at a rural clinic in Kenya between September 2018 and September 2019. The samples were then digitized using a portable scanner and uploaded through mobile networks to a cloud-based deep learning (DLS) system. Just under half of the smears were used to train the program to identify various preventable diseases while the rest were used to assess its accuracy.

The AI ​​assessment was then compared with that of the digital and physical samples by two independent pathologists. The study shows that the evaluations were very similar. The DLS had a 96-100 percent sensitivity in identifying patients with predictive injuries. Patients with more severe upper extremity injuries did not receive a false-positive evaluation. In terms of identifying lesions without lesions, the DLS performed the same assessment as the pathology in 78–85 percent of cases.

The researchers believe the method could be used to ban bulk smears, which would give local experts time to examine the ones that keep out smears. Before this can happen, however, larger and more diverse patient groups need more research, including more smears and different types of injuries as well as biopsies with a proven prognosis of cervical cancer.

“With the portable online microscope, the DLS can be a ‘significant helper’ when screening for breast cancer,” Lundin explains. “The AI ​​assistant can be accessed worldwide 24/7 and local experts can help diagnose many more smears. This approach will enable countries with limited resources to provide screening services to their population well. more efficient and at a lower cost than is currently the case. “

The study was conducted in collaboration with the Institute for Finland Molecular Medicine (FIMM) at the University of Helsinki, the Kinondo Kwetu Health Services Clinic in Kenya and Uppsala University in Sweden.

Two of the study’s authors, Johan Lundin and Mikael Lundin, are the founders and co-owners of Aiforia Technologies Oy, whose machine learning and image analysis platform was used in the study to develop the DLS. None of the other authors described conflicts of interest.