Detecting Pneumonia with Quantum AI

Pneumonia is one of the leading causes of death among young children and the elderly worldwide. It is often difficult to detect on X-rays, especially in the early stages. Automated image classification systems trained to distinguish healthy lungs from diseased ones can improve diagnostic accuracy. However, the medical image datasets used to train these models are often small and imbalanced in terms of the ratio of healthy to diseased individuals. This limits the development of robust classifiers with high accuracy.

A new quantum-based model developed at LMU’s Chair of Mobile and Distributed Systems can help diagnose diseases more quickly and accurately—for example, pneumonia on X-rays. In the future, this quantum-based system could achieve results comparable to those of similar classical approaches using only a fraction of the parameters required by classical models.

How It Works

Enlarge

The classification of medical image data is currently performed primarily using so-called neural networks, particularly convolutional neural networks (CNNs). Although these models achieve high accuracy, they carry the risk of “overfitting” when used with small datasets due to the large number of parameters that need to be optimized, and therefore often require the use of additional techniques such as transfer learning or regularization.

In a foundational study published last year, the researchers investigated the potential of quantum-based methods. The model they developed is based on so-called Quantum Boltzmann Machines (QBM), probabilistic models that learn probability distributions from data. The sampling process required for training and inference is carried out using quantum annealing, an optimization method that leverages quantum mechanical effects such as quantum tunneling.

The researchers then applied the method as part of a use case on the QuCUN quantum network platform, a collaborative project involving LMU, Aqarios, BASF, and SAP, funded by the Federal Ministry of Research, Technology, and Space. Using classified image data—in the current use case, chest X-rays of children from the MedMNIST dataset—the model learns a probability distribution across the data, in which relevant structural features, such as characteristic shadows or consolidations, are more likely to occur in patients with pneumonia than in healthy individuals. The model can then evaluate new, previously unknown images based on the learned features and assign the images to the “healthy” or “diseased” classes with a certain probability.

9,000 instead of 11 million trainable parameters

The results show that the QBM model achieves an accuracy of around 84–86 percent with fewer than 9,000 trainable parameters. While this falls short of established classical image classification models, which achieve about 94 percent on the same dataset, it uses only a fraction of the parameters. By comparison, a commonly used traditional CNN architecture such as ResNet-18 has more than 11 million trainable parameters.

Speed Gains Through Quantum Computing

As the study showed, quantum physics-based machine learning models—such as a Boltzmann machine based on quantum annealing—can, in certain cases, significantly reduce training times for image classification and identify complex feature correlations even in small datasets.

“Our research shows that quantum machine learning algorithms can offer specific advantages over comparable classical approaches—for example, when data availability is limited,” says Tobias Rohe, a doctoral student at the Chair of Mobile and Distributed Systems at LMU and a participant in the study. “The task now is to further explore these strengths, identify suitable use cases, and gradually transition the technology from research into everyday medical practice as quantum hardware matures. However, we must acknowledge that this is still a long—but all the more exciting—journey.”

Follow-up studies are therefore necessary to test the technology’s applicability to more clinically realistic datasets. The underlying quantum hardware and its practical implementation are also still in an early stage of development.

Daniëlle Schumann, Mark V. Seebode, Tobias Rohe, Maximilian Balthasar Mansky, Michael Schroedl-Baumann, Jonas Stein, Claudia Linnhoff-Popien, Florian Krellner: Quantum Boltzmann Machines Using Parallel Annealing for Medical Image Classification. IEEE Xplore 2025.

About QuCUN: QuCUNis Germany’s Quantum Computing User Network, a collaborative project between LMU, Aqarios, BASF, and SAP, funded by the Federal Ministry of Research, Technology, and Space. QuCUN helps industry partners prepare for the age of quantum computing—without requiring large upfront investments.