Single-formula symbolic classifier detects pediatric pneumonia on chest X-ray with AUC 0.89 on hold-out test

A closed-form symbolic classifier using radiomic features (entropy, solidity, fractal dimension) detected pediatric pneumonia on chest X-rays: AUC 0.93 under cross-validation, 0.89 on hold-out test, with external dataset re-calibration. Fully auditable; no black-box needed.

• Diagnostic-accuracy study; symbolic regression evolved a single non-linear equation from a compact radiomic feature set; evaluated by 10-fold cross-validation and a filtered independent hold-out set, plus external dataset re-calibration (design details and exact n not reported in source).
• Hold-out accuracy dropped to 79.1% vs. 87% at cross-validation — a gap worth noting; the hold-out set was 'filtered,' introducing potential selection bias, and no confidence intervals or p-values were reported.
• Model has far fewer parameters than competing deep learning architectures, making it a candidate for resource-constrained or regulatory-audit settings, but external prospective validation in diverse clinical populations has not yet been reported.

RadPigeon summaries are original and for information only. They are not clinical advice.