Diseases such as pneumonia, cardiomegaly, and hMPV have terrible health impacts around the world, especially in resource-limited environments that do not have expert radiologists. As a solution to the problem of needing faster and more accurate interpretable diagnostics, this study outlines a framework called Genetic Algorithm – Aided Deep Feature Selection (GA-DFS) which uses DenseNet121 for feature extraction and combines it with Genetic Algorithms (GA) to optimize feature selection on chest X-ray images. The methodology described utilizes the VinBigData dataset with over 18,000 annotated X-rays, performing image preprocessing, transfer learning, and hybrid neural network model training. In the proposed method, dense deep features extracted using dense neural networks are enhanced through GA refinement via selection, crossover, and mutation which leads to improvement in performance with lower computational costs. Results demonstrate that the GA + DenseNet model performed better over other architectures including DenseNet, ResNet50, VGG16, EfficientNetB3, achieving 86.75% classification accuracy (compared to the 83.47% of the DenseNet), while maintaining a lower inference time with minor concessions in F1-score and precision. The GA also enhanced the performance for models as VGG16 + GA (from 80.91% to 83.54%). The overall compile metrics also shows that DenseNet + GA model performs the best with 86.20%. This GA-based feature selection eliminated redundancies and overfitting and increased interpretability and efficiency broadening their use in clinical settings. This new model provides a stronger and more resilient respiratory disease diagnosis method. Future integration with electronic medical records, CT scans, and laboratory tests will form an extended web of multi-modal, context-aware evidence, enhancing diagnostic accuracy, minimizing false positives, and enabling risk-based patient stratification in clinical practice.