Evolution of histopathological breast cancer images classification using stochastic dilated residual ghost model


Abstract: Breast cancer detection is a complex problem to solve, and it is a topic that is still being studied. Deep learning-based models aid medical science by helping to classify benign and malignant cancers and saving lives. Breast cancer histopathological image classification (BreakHis) and breast cancer histopathological annotation and diagnosis (BreCaHAD) datasets are used in the proposed model. The study led to the resolution of four essential issues: 1) Addresses the color divergence issue caused by strain normalization during image generation 2) Data augmentation uses several factors like as flip, rotation, shift, resize, and gamma value in order to overcome overfitting concerns caused by a lack of histopathology images. 3) Using the proposed stochastic dilated convolution (SDC) model, able to find missing features and improve tiny and low-level features such as edge, contour, and color correctness. This model effectively solves depth issues with stochastic pooling and max pooling by combining three proposed units: spatial dilation convolution, channel dilation convolution, and dilated convolution to detect missing features and obtain more information on images. Stochastic pooling may choose any value since it is classified as a heterogeneous approach that can determine tiny and large values. When the max pooling and stochastic pooling are combined, a decent feature map is produced, processed by the dilatation unit, detecting cancer cells without adding to the complexity. 4) To correctly identify breast cancer and extract depth characteristics, the stochastic dilated residual ghost (SDRG) model uses the proposed SDC model, a ghost unit, stochastic upsampling, and downsampling units. The proposed feature selection unit employs linear transformations to produce feature maps to show information based on intrinsic characteristics to eliminate redundant or similar features in convolutional neural networks. Convolution and identity mapping are used to construct and retain intrinsic feature mappings in this skip unit while upsampling with stochastic pooling is used to minimize feature dimensions. The proposed model?s performance was assessed with 150,271 augmented and original histopathology images. The proposed method?s findings compare favorably to eight popular approaches. In addition, the suggested model outperforms the competition in terms of accuracy, average precision score, precision, sensitivity, and f1 score. With an accuracy of 98.41% with BreakHis and 98.60% with the BreCaHAD dataset with enhanced images, the proposed method exceeds numerous state-of-the-art methods.

Keywords: deep learning, stochastic pooling, dilation convolution, stochastic dilated residual ghost model, stochastic dilated convolution model, spatial dilated convolution unit, channel dilated convolution uni

Full Text: PDF