Invasive Ductal Carcinoma Detection In Breast Histology Images Using Dilated Depthwise Separable Convolutions

The second-leading cancer in women to be diagnosed is breast cancer. When the condition is identified early on, treatment for breast cancer can be very efficient, with mortality rates of 90% or higher. Using a variety of algorithms based on deep learning has boosted the efficacy of recognizing and detecting breast cancer. These breast image categorization methods, however, can generate an extensive number of training instances and render the model highly complex. To address this problem, this paper introduces an ultra - light neural network using Dilated Depthwise Separable Convolutional Network (DDSCNet) to minimize the network's attributes and computational burden throughout the convolution process. Additionally, dilated convolution is used to expand the receptive field while retaining the level of convolution parameters, allowing for the retrieval of additional high-level global semantic features to boost classification accuracy. The study's effectiveness is evaluated in terms of its accuracy, sensitivity, and specificity. The outcomes demonstrate that DDSCNet attained the maximum accuracy, sensitivity and specificity of 96.15%, 99.71%, and 99% respectively.