Mistakes and errors in healthcare can have significant consequences; therefore, accurate and precise text classification is particularly important in this field. Medical texts are highly specialized, with distinct vocabulary, terminology, and syntax, making them different from general texts. General text classification models often struggle to perform effectively in the medical field, exhibiting various challenges such as overfitting, reduced accuracy, and difficulty managing medical terminology. To address these issues, we propose a hybrid BERT-CNN model specifically designed for medical text classification. The approach involves three main steps: (1) medical text processing; (2) Text Representation based on BERT and (3) hybrid deep learning model to extract important features and text classification tasks. Using the cancer documents dataset, we conducted a comparative analysis and experiments, demonstrating that the proposed approach effectively mitigates the problems discussed above, achieving improved accuracy and robustness in handling medical terminology.