Dual-path Residual UNet with Convolutional Attention based Swin-Spectral Transformer Network for Segmentation and Detection of Ovarian Cancer

In this paper, a new transformer based deep learning network (TDLN) model has designed for detecting ovarian cancer (OC) from the input samples. Primarily, the input images are pre-processed to remove the unwanted noises and thereby enhances the quality through HSV color channel conversion, block matching with triple dimensional filtering (BM_TriD-Fil) and improved gabor wavelet transform (IGWT). From the pre-processed images, the affected regions are gets segmented through a new dual-path residual UNet (DPResU-Net) structure. Using the segmented regions, the feature extraction and classification is performed by utilizing a novel convolutional adaptive-attention based swin-spectral transformer network (CA
²
_S
²
TNet) model. In order to tune the parameters of CA
²
_S
²
TNet model, chaotic fire hawks optimization (CFHO) has employed. The proposedTDLN model is implemented in the python platform through TCGA-OV dataset and assessed the performance in terms of different evaluation measures. The maximum classification accuracy obtained by TDLN is 99.91%, superior to the existing approaches for OC classification.