Ensemble AI Model Classifies Lung Disease From CT Scans

A federated deep learning framework reports 98.96% accuracy in classifying COVID-19, pneumonia, tuberculosis and normal chest CT images.

Updated on Nov. 11, 2026 in Asthma

Large white CT scanner gantry centered in a quiet, softly lit clinical imaging room with pale walls and a single floor cable.
Researchers proposed a federated AI framework that classifies COVID-19, pneumonia, and tuberculosis from chest CT scans with reported 98.96% accuracy while keeping patient data local. AI Illustration. Upload story photo >

Researchers have proposed a federated IoMT-assisted ensemble deep learning framework for automated pulmonary CT analysis. The framework classifies COVID-19, pneumonia, tuberculosis and normal cases while collaborative training occurs across multiple healthcare institutions without directly sharing patient data.

Why it matters

The study targets the need for rapid, reliable and privacy-aware diagnostic tools for pulmonary diseases such as COVID-19, pneumonia and tuberculosis. Keeping patient data local while still training across institutions addresses a key barrier to building large clinical AI systems.

The reported performance figures come from testing on a multi-class chest CT dataset drawn from public sources, with 98.96% accuracy and 98.56% F1-score. Which comparison models were used and how the framework performs on unseen hospital data remain unreported.

The details

The framework combines an IoMT-enabled deployment architecture, Federated Averaging (FedAvg)-based distributed learning and an ensemble of pre-trained ResNet152V2, DenseNet201 and InceptionResNetV2 models. It uses transfer learning, Adam optimization, dropout regularization and early stopping to classify chest CT images.

Timeline

  1. The study's core event is dated November 11, 2026.

Health Landscape

The study applies the Federated Averaging (FedAvg) distributed learning strategy to chest CT diagnosis, extending a privacy-preserving training method to pulmonary disease screening. It reflects a broader push to use multi-institution AI training as an alternative to centralized patient-data pooling.

The framework is described as a potential decision-support system for pulmonary disease screening and remote healthcare. Its practical use for patients depends on clinical validation and deployment beyond the reported study.

The takeaway

Privacy-preserving federated training may let hospitals build stronger diagnostic AI without moving patient scans. Readers should watch for independent validation before such tools reach routine clinical care.

Further reading

For more on lung conditions, see the Asthma section.

Source note: This article includes information reported by Nature.