Breast cancer detection using edge detectors and deep learning

This research focuses on classifying thermogram images for breast cancer detection using a combination of edge detection methods and pre-trained deep learning models. Edge detectors such as Prewitt, Canny, Robert, and Sobel were applied to extract features from thermogram images, and the resulting edge-detected images were combined with original grayscale images to create 3-channel inputs. These inputs were then used to train classification models, including DenseNet201, DenseNet121, and a custom Keras model. To balance the classes of healthy and affected images, data augmentation techniques were employed, such as rotating healthy images to generate additional samples and equalize the number of images in both classes. The classification results showed varying performance based on the choice of edge detection methods and models, with certain combinations achieving improved accuracy, sensitivity, and specificity. The study emphasizes the importance of data preprocessing, augmentation, and exploring advanced feature extraction techniques like discrete wavelet transform to optimize classification performance.