Deep assisted dense model based classification of invasive ductal breast histology images

Invasive Ductal Carcinoma (IDC) is one of the types of breast cancer which is mostly diagnosed in the female. The diagnosis of IDC becomes a challenging task when a microscopic cell line structure is classified manually which requires a lot of expertise. Literature suggests a lot of computer-assisted deep models that prove as a helping hand for radiologists. But one limitation identified in most of the deep models is the vanishing gradient problem. This paper presents a dense model that overcomes the limitation of the vanishing gradient problem and outperforms in terms of classification accuracy with respect to other deep models. In addition, the proposed methodology also focused on the feature reusability with an optimized number of parameters for IDC histology image classification. The proposed method uses the CLAHE as a preprocessing method with 9 layers of dense deep architecture for experimentation purposes. The proposed model uses the base architecture of Densenet-169 with some proposed layer modifications to solve the targeted problem. While during the experimentation, validation of the proposed model is found satisfactory with an accuracy of 98.21%. The test set results are provided on the Kaggle under leaderboard with an AUC–ROC result of 97.54%. The experimentation result gained a classification accuracy of around 1.21% with respect to other methods present in the literature.