AI in mammography

Breast cancer is the most prevalent neoplastic disease in women and the leading cause of cancer death. Since the probability of recovery depends on the ability to diagnose cancer at an early stage, research aims to find innovative solutions to detect cancer at an early stage in screening programs.

In this context, we have undertaken several research projects in order to develop an automatic system based on artificial intelligence algorithms, capable of making predictions useful to radiologists as decision support during image analysis (QIDS).

This research project, developed in collaboration with the Politecnico of Turin, led to the creation of a software for the classification of mass-type tumor lesions in synthetic and tomosynthesis 2D images.

The methodological approach includes a first phase of pre-processing of the images and characterization of the lesion and of the breast tissue through the extraction of morphological/structural characteristics with an approach:

• Automatic, through deep learning algorithms that involve the use of convolutional neural networks (CNN);
• Manual, using quantitative imaging techniques.

The second phase is dedicated to the construction of predictive models to identify the class of lesions (benign or malignant) through the following Machine Learning methods:

• Linear Discriminant Analysis (LDA);
• Support Vector Machine (SVM);
• K-Nearest Neighbors (KNN);
• Artificial Neural Networks (ANNs).