A Robust DL Approach for Detection of Invasive Ductal Carcinoma in Whole Slide Images using DenseNet169

The present research introduces a deep learning method for automatically recognizing and evaluating tissue areas that correspond to invasive ductal carcinoma (IDC) in whole slide images (WSI) of breast cancer (BCa). Learn-from-data techniques which utilize computational simulations of the learning process are called deep learning approaches. It has been demonstrated that these methodologies outperform conventional methods in resolving the challenging issues in a number of domains, particularly object and speech recognition. The main reason why making an invasive breast cancer evaluation is an exhausting and laborious process is that, it necessitates a pathologist to scan a huge number of benign regions with the aim to determine the spots that are malignant. Effectively characterizing IDC in WSI is a prerequisite for evaluating the severity of tumors and predicting patient prognosis. In order to support diagnosis, the DL approach in this study draws out many convolutional neural networks (CNNs) for perceptible semantic examination of tumor locations. A WSI database of 162 patients with IDC diagnoses is exploited to assess the technique, and 170 slides were chosen for training. The objective of the experimental analysis was to assess the accuracy of the classifier in identifying IDC tissue areas in WSI. A fusion of optimization techniques namely Best First and Evolutionary Search has been employed for feature reduction and improvement. When using DenseNet169, our approach produced the highest quantitative outcomes for the computerized identification of IDC sections in WSI with regard to accuracy 87.46% in comparison with an approach using EfficientNetB0 and ML classifiers for invasive tumor categorization employing Random Forest, BayesNet, SVM, Naïve Bayes and Decision Tree with accuracy of 80.65%.