• Automatic pancreatic cancer segmentation and classification based on hybrid spatial-channel transformer with lightweight BiLSTM

    G. Raja & U. Srinivasulu Reddy 

    Abstract

    The scenario is depressing for pancreatic cancer, a deadly form of the disease. Automated pancreatic cancer segmentation and classification using a Deep learning (DL) method is used to track, predict, and classify the presence of cancer. Deep learning algorithms can also offer accurate image analysis for comprehensive diagnostic information. This paper proposes a novel approach for automatic pancreatic cancer segmentation and classification using deep learning techniques. Initially, input images are segmented by using label-decoupled network with spatial-channel transformer and attention (LDNet-SCTrans-Att). Next, shape and texture features are extracted using pyramidal histograms oriented gradient and gray level run length matrix (PHOG-GLRLM). The Depthwise Separable Dilated Convolutional Bidirectional long short-term memory (DSDC-BiLSTM) network is then used to classify the pancreatic cancer images. Furthermore, classifying parameters are fine-tuned by using hybrid genetic dung beetle optimizer (HGDBO) algorithm. As a result, the proposed method is compared to other existing deep learning methods. Also, the proposed method is analyzed using python tool and the results are evaluated using kaggle datasets. The obtained simulation results prove that the proposed method provides better results in term of accuracy (99.69%), precision (99.48%), recall (99.53%), f1-score (99.51%) and specificity (99.68%).