Automated classification of ganglions and spiculated masses: A case study, Mexico's National Institute of Cancerology
Sergio Lenis, Cristián Castillo-Olea, Begonya García-Zapirain, Eric Ortiz · 2019
The proposed technique is based on Mask R-CNN and was evaluated using INCAN data set. We obtained a ganglion and spiculated mass average precision rate of 43% and pixel-wise segmentation performances with 69% recall, 99% specificity, 64%precision, 99% accuracy, and 75% F1score. The data set is validated by 5 experts: oncologist and specialist in mammary pathology, three radiologists and specialists in mammary pathology, and a surgeon mastologist. This approach allows us to employ deep learning techniques to provide assistance to health-care professionals in the medical diagnosis of breast cancer.