Hybrid dual attentional-based pyramidal deep learning model for liver tumor detection

Liver tumors are life-threatening diseases that occur due to the rapid growth of liver cells and even result in a chronic condition. Nevertheless, conventional liver tumor detection struggled with the limitations of poor robustness, difficulties in identifying the complex boundaries, and inconsistent segmentation. Hence, the research introduced a hybrid dual attention-based pyramidal deep convolutional network and deep belief network (HDPCBN) for accurate liver tumor detection. Significantly, the HDPCBN approach deployed a pyramidal architecture to learn deep hierarchical features and intricate patterns of the tumor effectively. Moreover, the integration of the dual attention technique gathers the long-range relationships of both spatial and channel dimensions for accurate liver tumor detection. In addition, the HDPCBN model learns the local patterns and edges from the images, thereby leading to precise tumor detection. The Hybrid deep flow features (HDF2) extraction gathers the discriminative and textural features of the tumor, which significantly reduces the training complexity and leads to accurate tumor detection. Experimental results indicate that the HDPCBN framework obtains exceptional performance, reporting the specificity of 93.77%, accuracy of 96.52%, sensitivity of 97.89%, balanced accuracy of 95.83%, and Macro F1-score of 96.67%, while outperforming the other state-of-the-art methods.