Calib MOPSO: A Calibration Aware Multi Objective Feature Selection Framework for Pneumonia Detection from Chest X Ray Images

Pneumonia is a leading global health concern, and chest X-ray imaging is widely used for its diagnosis. Manual interpretation of radiographs is time-consuming and subject to variability, motivating the need for automated and reliable computer-aided systems. We propose Calib-MOPSO, a calibration-aware multi-objective feature selection framework for pneumonia detection. Deep features are extracted using a fine-tuned ResNet-۵۰ with a ۵۱۲-dimensional projection head, and feature subsets are optimized through a multi-objective particle swarm optimization procedure. The framework jointly maximizes discrimination (average precision), promotes sparsity, improves probability calibration (Brier score). Selected features are classified with a probability-calibrated support vector machine using a radial basis function kernel. Experiments on the benchmark chest X-ray dataset show that Calib-MOPSO reduces features by nearly ninety-four percent while achieving strong discrimination (area under the ROC curve of ۰.۹۹۸ and area under the precision-recall curve of ۰.۹۹۹). On the held-out test set, the method maintained high recall and generalized well to Malaria and COVID-۱۹ datasets. Overall, Calib-MOPSO yields compact, calibrated, and generalizable models suitable for medical imaging tasks.

Keywords:

Pneumonia detection , chest X-ray , multi-objective optimization , feature selection , calibration , stability