Multiple Scaling Based EfficientNet Modelling for Liver Tumor Classification on CT Images

For both women and men over 60, liver cancer is the primary cause of cancer-related deaths. To help physicians to diagnose patients more accurately, computer-assisted imaging techniques have become increasingly important in recent years. Recent, deep Convolutional Neural Network (CNN) research has produced amazing improvements in image segmentation and classification. The same issue of diagnosing liver nodules in computed tomography (CT) scans is addressed in this research by introducing a novel Computer-Aided Detection (CAD) system that makes use of an Efficient Network (EfficientNet) image classification algorithm. Unlike CNN, which adjusts its network parameters arbitrarily, a set of predetermined scaling coefficients is used in the EfficientNet scaling technique to reliably scale the network’s breadth, depth, and resolution. Here the EfficientNet models are assessed by varying the input dimensions of the CT scans from The Liver Tumor Segmentation (LiTs) dataset. Finally, the performance evaluation shows that the input dimension 224×224 effectively classified the images and is superior to the other models evaluated with 0.991 AUC and 99.37% F1-Score, precision 99.44%, recall 99.30%, specificity 99.43%, and accuracy 99.36% for Kaggle datasets.