Classification Performance of BFGS Quasi-Newton Backpropagation and Scaled Conjugate Gradient Backpropagation Models for Lung X-Ray Images

The lungs are vital organs essential for human survival, and their health requires significant attention. Common lung disorders, such as pneumonia and COVID-19, have caused numerous fatalities, underscoring the need for rapid and accurate diagnostic tools. This study evaluates the performance of two optimization algorithms, BFGS quasi-Newton backpropagation (TrainBFGS) and scaled conjugate gradient backpropagation (TrainSCG), in training Multi-Layer Perceptron (MLP) models for this task. A dataset consisting of 1,184 lung X-ray images was used, with features extracted using Gray Level Co-occurrence Matrix (GLCM). The MLP model was trained and tested with 5 and 15 hidden nodes to determine the optimal configuration. The results show that the TrainSCG model generally outperforms the TrainBFGS model, achieving the highest average training accuracy at hidden node 5 is 77.97%, while for hidden node 15 the TrainBFGS model outperforms than TrainSCG with training accuracy 78.53%. Based on the running results at hidden nodes 5 and 15, the TrainBFGS model outperforms the TrainSCG model. TrainBFGS can obtain the highest training accuracy on the first run for hidden node 5 and the eighth run for hidden node 15. While TrainSCG only obtained the highest training accuracy on the tenth run on hidden nodes 5 and 15.