Beyond the colors: enhanced deep learning on invasive ductal carcinoma


Neural Computing and Applications, vol.34, no.21, pp.18953-18973, 2022 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 34 Issue: 21
  • Publication Date: 2022
  • Doi Number: 10.1007/s00521-022-07478-w
  • Journal Name: Neural Computing and Applications
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Applied Science & Technology Source, Biotechnology Research Abstracts, Compendex, Computer & Applied Sciences, Index Islamicus, INSPEC, zbMATH
  • Page Numbers: pp.18953-18973
  • Keywords: Convolutional neural networks, Invasive ductal carcinoma, Hematoxylin, Eosin staining, Color space, BREAST-CANCER, IMAGE-ANALYSIS, CLASSIFICATION
  • Kayseri University Affiliated: No


© 2022, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.Breast cancer is one of the most common cancer-related causes of morbidity and mortality in women around the world, and its early detection is essential for successful treatment. Invasive ductal carcinoma (IDC) is the most prevalent phenotypic subtype of all breast cancers, accounting for 80% of all cases. Deep learning (DL) has been used to diagnose breast cancer in a variety of situations. Our aims with this study were to contribute early detection of IDC as well as other cancer types and to help physicians while inspecting whole slide images belonging to possible cases. This study will also explore ways of improving the performance of deep learning models. While evaluating the models, the Matthew correlation coefficient (MCC) score was preferred instead of the F1 score, which did not take into account the real negatives. Pretrained networks (Xception, InceptionResNetV2, NASNetLarge) and two custom models, feature extraction technique, various color spaces (RGB, Gray, CMYK, HSV, HED, YIQ, YUV, YCrCb, LAB, LUV, XYZ), color changes (square, polynomial, multiplicative inverse) and ensemble modeling were tested. Our findings show that color data in image patches are crucial. The performance improvement of the convolutional neural network (CNN)-based model is achieved by generating various color spaces and their changes from the input channels in the preprocessing stage. It also showed that, despite the computation time advantage, feature extraction reduced the MCC score. Our novel ensemble methods based on precision and convex hull density were tested. We made an ensemble model with better MCC(0.6931) and F1(0.7958)scores on test data, while the first study on this dataset has an F1 score of 0.7180.