Assessment of the Breast Cosmesis Using Deep Neural Networks: an Exploratory Study (ABCD)

A new algorithm was introduced which is based on deep neural network (DNN) which receives an image as input and returns the coordinates of the breast key points as output. These key points are then given to a shortest-path algorithm that models images as graphs to refine breast key point localization. The algorithm learns, directly from the image, to compute features and to use those features in the analysis of the aesthetic result. This comprises of two main modules: regression and refinement of heatmaps, and regression of key points. To perform the heatmap regression, the U-Net model is used.

The goal of the first module is to generate an intermediate representation consisting on a fuzzy localization for the key points that are to be detected.

The second module receives and refines this fuzzy localization, and through complex calculations, outputting the x and y coordinates of the keypoints, and the data generated from which can be used for disease / image classification.

Inclusion criteria

• Confirmed diagnosis of primary breast cancer (invasive or in situ)
• Patient undergone breast conservation / Whole breast reconstruction
• Patient received breast RT
• Already provided written informed consent on earlier projects
• Patient provided photographs of both breasts
• Non-metastatic disease or oligometastatic
• Age > 18 years
• Reconsent given

Exclusion criteria

• Mastectomy without whole breast reconstruction
• Bilateral breast cancer
• Partial breast irradiation
• Male patient
• Limited life expectancy due to co-morbidity
• Patients undergoing brachy boost