Enhancing Liver Cancer Diagnosis with Hybrid CNN-Transformer Architectures

Diagnosing liver cancer at an early stage is crucial to enhancing the possibility of treatment success and patients' survival rates because it remains a significant global health problem. This paper proposes a new CNN-Transformer model for diagnosing liver cancer from medical photos for better hospital diagnosis. The proposed model integrates feature extraction in the spatial domain with the help of CNNs while combined with the global context-capturing ability of Transformers, meaning that it removes the constraints of single models. Specifically, the hybrid model was trained and tested by a liver cancer image dataset that is available to the public. It was also compared with solo CNN and Transformer devices. The precisions achieved by the hybrid model were 95 per cent. Hence, we achieved an accuracy improvement of 2% and significantly outperformed the CNN (88. 7%) and Transformer (90. 1%) models. The hybrid model got the accuracy and the recall value of 94 per cent. 5% and 96. 0%, respectively. In this regard, the CNN model was established to offer a precision of 89 per cent of its outcomes, compared to an overall increase. 2% and a recall of 87. 5% the Bleu-4 score is 95%, while the Transformer model is 91%. 3 % and the average number of relevant documents recalled was 89—8%. The result evaluated on the chosen F1-score shows the model's obvious effectiveness of 95%. 2% surpassed the score of CNN 88%. It stands for 3 %, and Transformer's score is 90.5%.