In this paper, we propose a novel method named SSL-mC (Self-supervised Learning with fine-tuning and meta-Classifier integration) for improving chest X-ray image classification. SSL has emerged as a powerful solution to address the scarcity of labeled medical images by enabling representation learning from unlabeled data. Instead of relying solely on fine-tuning a self-supervised pretrained model, the proposed strategy involves training nonlinear classifiers on the output probabilities of fine-tuned models across five architectures: ResNet-50, DenseNet-121, MobileNet-v2, Vision Transformer, and Swin Transformer. These output probabilities are concatenated to form input features for classifiers. The approach benefits from model-level integration to overcome the limitations of single-model and improve performance. Experimental results demonstrate that our proposed method outperforms fine-tuning approaches, with accuracy improvements ranging from 1.9% to 3.9% over the baseline models, achieving a highest accuracy of 88%.