A Heterogeneous Ensemble for Multi-Center COVID-19 Classification from Chest CT Scans

TL;DR

This paper introduces a diverse ensemble of nine models for robust COVID-19 classification from chest CT scans across multiple hospital centers, addressing domain shift and overfitting issues to improve diagnostic accuracy.

Contribution

The study presents a novel heterogeneous ensemble combining three inference paradigms and advanced training techniques to enhance multi-center COVID-19 classification performance.

Findings

Achieved an average macro F1 score of 0.9280 across four hospital centers.

Outperformed the best single model by +0.031 F1 score.

Reduced overfitting significantly through specialized loss and augmentation methods.

Abstract

The COVID-19 pandemic exposed critical limitations in diagnostic workflows: RT-PCR tests suffer from slow turnaround times and high false-negative rates, while CT-based screening offers faster complementary diagnosis but requires expert radiological interpretation. Deploying automated CT analysis across multiple hospital centres introduces further challenges, as differences in scanner hardware, acquisition protocols, and patient populations cause substantial domain shift that degrades single-model performance. To address these challenges, we present a heterogeneous ensemble of nine models spanning three inference paradigms: (1) a self-supervised DINOv2 Vision Transformer with slice-level sigmoid aggregation, (2) a RadImageNet-pretrained DenseNet-121 with slice-level sigmoid averaging, and (3) seven Gated Attention Multiple Instance Learning models using EfficientNet-B3, ConvNeXt-Tiny, and EfficientNetV2-S backbones with scan-level softmax classification. Ensemble diversity is further enhanced through random-seed variation and Stochastic Weight Averaging. We address severe overfitting, reducing the validation-to-training loss ratio from 35x to less than 3x, through a combination of Focal Loss, embedding-level Mixup, and domain-aware augmentation. Model outputs are fused via score-weighted probability averaging and calibrated with per-source threshold optimization. The final ensemble achieves an average macro F1 of 0.9280 across four hospital centres, outperforming the best single model (F1=0.8969) by +0.031, demonstrating that heterogeneous architectures combined with source-aware calibration are essential for robust multi-site medical image classification.