Optimization of Matching Networks with Transfer Learning in Few-Shot Pneumonia Detection

Pneumonia remains the leading cause of death among children under five years of age, with approximately 1.6 million deaths annually. Early detection is the key to reducing child mortality. However, most of the traditional diagnostic methods depend on the availability of trained personnel and medical resources, which are particularly limited in low-resource settings. While machine learning has provided a promising technology for early detection of pneumonia, its uses often suffer from the problem of a scarcity of labeled data needed to train robust models. In this study, we propose an optimized model for one-shot pneumonia detection that incorporates transfer learning with the matching networks. The proposed model utilizes a pre-trained MobileNetV3 model for feature extraction to produce high-quality embeddings that Matching Networks can use to classify pneumonia instances using a minimal number of labeled examples. The experimental results revealed that the proposed model outperformed state-of-the-art traditional machine learning algorithms such as random forest and support vector machines with a high accuracy of 93.21%, precision of 93.34%, recall of 93.20%, and F1 score of 93.19%. The proposed model showed relatively competitive performance compared to CNNs by attaining AUCs of 1 for COVID cases, 0.98 for normal cases, and 0.98 for pneumonia. These results indicate that the proposed model effectively balances classification performance with data efficiency and, as such, can be effectively deployed in resource-constrained environments.