Heart disease is a major cause of mortality worldwide, making early and accurate diagnosis essential. A multimodal framework for heart disease classification is presented through the integration of electrocardiogram (ECG) signals and clinical data. ECG recordings are obtained from the PTB-XL dataset, which includes clinical attributes such as age, gender, and diagnostic labels. A Residual Neural Network (ResNet) is employed to extract discriminative features from ECG signals, while Bidirectional Encoder Representations from Transformers (BERT) is utilized to encode clinical text data and capture contextual dependencies. The extracted features are fused using a fully connected architecture to enhance classification performance. Experimental results demonstrate an accuracy of 96.77%, indicating improved performance over unimodal approaches and supporting reliable clinical decision-making.