A research team from Innopolis University, LETI (Saint Petersburg Electrotechnical University), and Jadavpur University has developed an artificial intelligence model that helps detect breast cancer with a high degree of reliability. Instead of using a traditional approach, the researchers combined two compact neural networks — SqueezeNet and ShuffleNet. Through a mutual gating mechanism, the models exchange data, identify key signs of disease in tissue images, and filter out irrelevant information.
The model was tested on international histopathological image datasets. When identifying normal tissue, benign tumors, and different stages of cancer, it achieved an accuracy of 97%. Detection of malignant tumors reached 99% accuracy, while on a reference dataset with 100× magnification, the model achieved up to 99.72%.
The authors believe that this architecture could help regional hospitals and local healthcare facilities that do not have access to expensive medical technologies.