Invasive Ductal Carcinoma Detection using Convolutional Neural Network Architecture from Breast Histopathology Images

The purpose of the work is to create a robust convolutional neural network-based classification model to distinguish images of invasive ductal carcinoma (IDC) breast disease. This collection includes histopathological images from breast tissue samples some with IDC-positive and others IDC-negative. First in data preparation are resizing images to 50x50 pixels and arranging them for training and testing. The sample consists of 3,747 IDC-positive images and 21,031 IDC-negative images for 24,678 images overall. The model's CNN architecture consisted of many layers: convolutional layers, max-pooling layers, batch normalization, and dropout layers aimed at reducing overfitting. Trained with binary cross-entropy loss function and an Adam optimizer running at 0.0001 learning rate. After 40 iterations on the training set, the model's accuracy on the testing set was 94.73%, and on the training, set was 99.12%. The performance of the model was assessed using several parameters including accuracy, precision, recall, and F1-score. The categorization report shows that both IDC-positive and IDC-negative classes have excellent recall and accuracy. Moreover, confirming the ability of the model to classify IDC-positive and IDC-negative samples is the confusion matrix. This unique CNN-based classification model of breast histomorphology images seems to do remarkably well in identifying IDC, showing both excellent accuracy and the capacity to distinguish between positive and negative IDC samples. The model could enable pathologists to identify breast cancer, therefore enhancing the accuracy and effectiveness of breast cancer screening campaigns. Future directions of this work could concentrate on improving the performance of the model with larger datasets and investigating alternative image augmentation techniques that will generalize its capabilities on diverse input datasets.