<p>Hepatic tumor segmentation from Computed Tomography (CT) images relies on intelligent computer-aided algorithms for early detection and precision diagnosis. Combining multi-level learning concepts improves the precision of segmentation and detection from any image type. In this article, a Hybrid Duo-Transformer Network (HDTN) is designed to improve the precision of hepatic tumor segmentation from any quality of CT input. The proposed network model assimilates a self-attuned transformer and cascaded dilated layer for improving the segmentation accuracy. This hybrid model is designed to eradicate the problem of pixel dissolving issues in low-quality CT images. By extracting the intensity and saturation features, the transformer network identifies the segmentation boundary using a cross pixel present in the dissolving region. For this purpose, this network requires a minimum of 1 × 1 block of the input image. Following this boundary detection, the dilated layer is responsible for verifying the intensity within the least possible block region. The high-intensity outcomes are further extracted to converge the segmentation region. The failing intensities due to saturation are used to train the convolution layer to reduce the transboundary error pixels. Thus, the converged final block is the segmented region from the CT input with high precision. The proposed model improves the segmentation precision by 11.51% and intensity analysis by 8% and reduces the mean error by 9.09% for the different boundaries.</p>