Preprocessing-Induced Ranking Reversal is Architecture-Dependent in Cross-Domain Pneumonia Detection

Abstract

This study investigates how preprocessing selection influences the cross-domain generalization of convolutional neural networks (CNNs) for pneumonia detection and whether it changes performance rankings among architectures. Five preprocessing conditions—Baseline, Standard Deviation Normalization, Contrast Limited Adaptive Histogram Equalization (CLAHE), Histogram Equalization, and StdDev+CLAHE—were combined with three architectures: a simple CNN, ResNet-50, and EfficientNet-B0. The resulting 15 models were trained on the public Kaggle dataset containing 5,863 chest X-ray images and externally evaluated on 1,800 clinical chest X-rays collected in Central Java, Indonesia. Results showed that preprocessing rankings were architecture-dependent. The simple CNN achieved its highest external Recall with Histogram Equalization (0.9856), although Baseline ranked first internally. ResNet-50 was most stable with Standard Deviation Normalization, which ranked first internally (F1 = 0.9821) and externally (Recall = 0.9944). EfficientNet-B0 achieved the highest external Recall with CLAHE (0.9994), but its lower internal precision (0.77–0.87) indicated bias toward the pneumonia class. These findings demonstrate that high benchmark performance may not transfer directly to regional clinical settings because preprocessing interacts with scanner characteristics, data distribution, and network architecture. Therefore, each preprocessing–architecture pair should be validated on locally acquired images before clinical deployment.

ICU-acquired pneumonia beyond pathogen detection
Ignacio Martin-Loeches, Antoni Torres
Nature Communications16 Oct 2026Peer-reviewedCC BY-NC-ND 4.0Full textGeneral medicineCrossref
Physiologically Informed Digital Auscultation for Pneumonia Detection in Long-term Care Residents
Nicholas Rasmussen, Oleg Zaslavsky, Zih-Ling Wang +5
arXiv23 Sep 2026PreprintCC BY 4.0Full textPhysiologyarXiv
Concordance of next-generation sequencing and conventional microbiology for pathogen detection and antimicrobial resistance prediction in pneumonia
Nurlan Sandybayev, Radmir Sarsenov, Vitaliy Strochkov +5
Royal Society Open Science30 Sep 2026Peer-reviewedCC BY 4.0Full textGeneral medicineCrossref
Stroke-associated pneumonia after acute ischemic stroke: early temporal dynamics, nonlinear risk factors, and time-dependent prognostic impact
Shitong Zhang, Qiaomei Yan, Zheng Wang +4
Frontiers in Neurology20 Oct 2026Peer-reviewedCC BY 4.0Full textNeurologyCrossrefEurope PMC
In-race continuous glucose monitoring in highly trained male distance runners: a context-dependent reversal of the glucose–performance association between training and competition
Taira Kajisa, Toshiyuki Sakai
Scientific Reports8 Oct 2026Peer-reviewedCC BY 4.0Full textSurgery & anaesthesiaCrossref

Cite

BibTeX RIS

BibTeX

@article{sanjaya2026preprocessinginduced,
  title = {{Preprocessing-Induced Ranking Reversal is Architecture-Dependent in Cross-Domain Pneumonia Detection}},
  author = {Ucta Pradema Sanjaya and Sri Mujiyono and Ade Pratama and Iwan Setiawan Wibisono},
  journal = {Journal of Information Systems and Informatics},
  year = {2026},
  month = oct,
  doi = {10.63158/journalisi.v8i5.1944},
  url = {https://doi.org/10.63158/journalisi.v8i5.1944},
  note = {CC BY},
}

RIS

TY  - JOUR
TI  - Preprocessing-Induced Ranking Reversal is Architecture-Dependent in Cross-Domain Pneumonia Detection
AU  - Ucta Pradema Sanjaya
AU  - Sri Mujiyono
AU  - Ade Pratama
AU  - Iwan Setiawan Wibisono
PY  - 2026
DA  - 2026/10/24
JO  - Journal of Information Systems and Informatics
DO  - 10.63158/journalisi.v8i5.1944
UR  - https://doi.org/10.63158/journalisi.v8i5.1944
AB  - This study investigates how preprocessing selection influences the cross-domain generalization of convolutional neural networks (CNNs) for pneumonia detection and whether it changes performance rankings among architectures. Five preprocessing conditions—Baseline, Standard Deviation Normalization, Contrast Limited Adaptive Histogram Equalization (CLAHE), Histogram Equalization, and StdDev+CLAHE—were combined with three architectures: a simple CNN, ResNet-50, and EfficientNet-B0. The resulting 15 models were trained on the public Kaggle dataset containing 5,863 chest X-ray images and externally evaluated on 1,800 clinical chest X-rays collected in Central Java, Indonesia. Results showed that preprocessing rankings were architecture-dependent. The simple CNN achieved its highest external Recall with Histogram Equalization (0.9856), although Baseline ranked first internally. ResNet-50 was most stable with Standard Deviation Normalization, which ranked first internally (F1 = 0.9821) and externally (Recall = 0.9944). EfficientNet-B0 achieved the highest external Recall with CLAHE (0.9994), but its lower internal precision (0.77–0.87) indicated bias toward the pneumonia class. These findings demonstrate that high benchmark performance may not transfer directly to regional clinical settings because preprocessing interacts with scanner characteristics, data distribution, and network architecture. Therefore, each preprocessing–architecture pair should be validated on locally acquired images before clinical deployment.
SN  - 2656-5935
SN  - 2656-4882
ER  -

Abstract

Related

Cite