Pancreatic Cancer Prediction Using Hierarchical Convolutional Neural Network and Visual Geometry Group16 CNN Approach on Accuracy and Specificity Performance Measures

Abstract

The main aim of the work is to measure the accuracy, sensitivity, and specificity in the detection of pancreatic cancer. Materials and methods: The dataset of 7390 images obtained from Kaggle for training (80%) and testing (20%) of the predictive model developed in Matlab and statistical analysis is done using SPSS software. The hierarchical convolutional neural network (HCNN) is used for developing the model and compared with the visual geometry group (VGG16) based model. Result: The predictive model developed using the HCNN algorithm shows an improvement in accuracy 94.8221.4705, specificity 88.49801.78406, and sensitivity 90.04601.79226 than VGG16 model with accuracy 94.3760, specificity 88.39200.1663, and sensitivity of 88.92600.5808 with the significance of 0.437, 0.497, and 0.165. Conclusion: The outcome of the study confirms that the HCNN based model appears to be of higher accuracy than the VGG16 based model.