How to Enhance Mammogram Analysis with AI?

The Institute for Artificial Intelligence of Serbia, in collaboration with the Institute for Oncology and Radiology of Serbia, is developing a software application to improve the quality of mammogram analysis using artificial intelligence models.

“On a technical level, deep learning methods, particularly convolutional neural networks, which are often applied in image processing, are used. After defining the architecture of the neural network, including layers and neurons, the model is trained on a large number of labeled images. After this ‘learning’ process, the model can recognize changes in tissue in mammograms with a certain probability. Thus, our model would provide probability estimates for lesions in breast tissue and rank mammogram images, giving recommendations to specialist doctors on the order in which to review them. The basic idea is to assist often overburdened radiologists by optimizing their time”, says Dr. Vanja Švenda, a research associate at the IVI and a member of the “Computer Vision & Remote Sensing” research group, in an interview for Dnevnik.rs.

Junior research assistant Nikola Jovišić has been involved in the project from the beginning and emphasizes that from the conception of the idea, data collection, model development, to its implementation, artificial intelligence will not independently establish diagnoses or make decisions about further patient treatment.

“This is a model that only provides recommendations to doctors and should be viewed as such, as a tool whose task, in this specific case, is to support medical staff in making final decisions. The goal is primarily to avoid false-negative findings. AI enables the rapid processing of a large number of images, significantly reducing the time needed for diagnosis. By triaging patients and ranking images according to the likelihood of lesions, doctors can focus on the highest-risk cases. Finally, AI models have the ability to continuously learn and improve based on new data, which further increases the accuracy and efficiency of diagnostics”, Jovišić notes.

Read more about how artificial intelligence can assist Serbian radiologists at the following link.