Quantum Model Shows Promise in Lung Cancer Classification

QUANTUM-assisted machine learning could help classify lung cancer when training data are limited, according to the results of an exploratory computational study.

The best-performing approach achieved 91.90% accuracy on a real quantum computer. However, the findings come from a small, augmented dataset and should be interpreted as an early technical demonstration rather than evidence of readiness for clinical diagnosis.

Learning From Limited Records

Researchers began with 309 records containing demographic information, symptoms and lifestyle factors. They used synthetic minority oversampling to create a balanced dataset of 540 records, with equal numbers labelled as lung cancer and non-cancer.

A feature-selection method reduced the inputs to 10 variables. Three quantum-assisted classifiers were then compared across different computing frameworks, using 108 samples for training and the remaining 432 for evaluation.

The strongest results came from a quantum support vector machine implemented through the D-Wave framework. Accuracy reached 94.31% in simulation and 91.90% on quantum hardware. The authors presented the findings as a possible route towards decision-support systems that can learn from relatively limited data. Springer Nature Link

Understanding the Accuracy Figures

Accuracy describes the proportion of classifications that agree with the dataset labels. It does not, by itself, establish how well a model would perform in a screening population or among patients presenting with uncertain symptoms.

The use of synthetic records also requires careful interpretation. Increasing the number of samples through augmentation does not provide the same evidence as recruiting additional, independent patients.

A further methodological concern is the sequence of preprocessing and evaluation. The reported workflow describes oversampling before subsequent model assessment. This creates a potential risk that related information could appear across training and evaluation samples, making independent validation especially important.

These considerations do not invalidate the technical comparison, but they limit what can be concluded about clinical performance.

Further Validation Needed

The next question is whether the approach can maintain useful performance in genuinely independent clinical datasets. Such evaluation would need to examine missed cancers, false-positive classifications and performance across different patient groups.

A fair comparison with well-tuned conventional models would also help establish whether quantum computing offers a practical advantage.

For clinicians, the study is best viewed as an investigation of an emerging computational method. It does not show that the system can replace imaging, pathological assessment or established diagnostic pathways. Its potential value will depend on whether promising laboratory results translate into reliable and useful decisions in clinical care.

Reference

Hossain MM, Himal MHH, Munir A. Inferring lung cancer under limited learning data through gate-based and annealing-based quantum techniques. Discover Quantum Science. 2026;2:34. doi:10.1007/s44464-026-00038-4.