<p>Currently, liver tumors are the major cause of cancer death. Exact evaluation and segmentation of liver tumors are significant to suggest appropriate treatment and to monitor the progression of the tumor. However, traditional diagnosis models “like Magnetic Resonance Imaging (MRI) and Computed Tomography (CT),” consume more time since the physician has to examine the tumor by employing 3-dimensional Computed Tomography (CT) images, which comprise multiple lesions. Recently, techniques based on artificial intelligence have offered more accurate outcomes. Yet, the real effect of these approaches on physicians, when employed in the medical network is still undefined. Moreover, diagnosis of a liver tumor is hard, because of the occurrence of noise, varying nature, and non-homogeneity. Therefore, an “intelligent liver tumor segmentation and classification framework” is introduced based on deep learning. In this proposed method, “initially, the necessary images are garnered from the benchmark datasets.” Further, the garnered images are given to the developed Transformer-based Dense Recurrent Residual Unet++ (Trans-DR2Unet++) model to segment the abnormalities in the collected liver tumor images. Here, the developed Trans-DR2Unet++ integrates transformer mechanisms to enhance feature extraction from complex medical images, enabling more accurate delineation of structures and pathologies. Later, segmented images are fed into Pyramid Dilated-Adaptive Residual DenseNet (PD-ARDNet) model for classifying liver tumors. The proposed PD-ARDNet Network leverages the strengths of Residual DenseNet technique to create powerful and flexible network architecture. Here, an Improved Arbitrary Variable-based Secretary Bird Optimization (IAV-SBO) is applied to improve the “classification efficiency by tuning the hyper-parameters” of PDRDNet-AT. Finally, numerical experiments are carried out for the developed model by employing several performance metrics to ensure the effectualness of the suggested work. Optimizing the classification process enhances model performance, efficiency, and reliability. The integration of IAV-SBO with PD-ARDNet not only enhances classification but also optimizes computational efficiency. After the classification process, the classified images are the outputs for this proposed method. As a result, this technique leverages the strengths of both precise image segmentation and robust classification, providing enhanced accuracy in identifying liver tumors. The proposed model’s accuracy is 7.99%, which is better than other existing techniques. The accuracy and reliability of this method can significantly improve patient outcomes by facilitating earlier diagnosis and more tailored treatment plans. The results proved that the developed liver tumour diagnosis using Trans-DR2Unet++ for Segmentation and PD-ARDNet-based classification technique, resulting in enhanced accuracy and efficiency in segmenting and classifying tumors. The Trans-DR2Unet++ helps to precisely segment liver tissues and tumours, provides a strong foundation for subsequent analysis, while PD-ARDNet’s advanced classification capabilities enable the system “to accurately identify different types of liver tumours.” The overall framework provides a comprehensive solution for liver tumor diagnosis, minimizing false positives and enhancing clinical decision-making.</p>