Inferring lung cancer under limited learning data through gate-based and annealing-based quantum techniques

Abstract

Early diagnosis of lung cancer and reliable decision support are crucial steps since this is a leading cause of cancer-related mortality worldwide. This research aims to build an approach using quantum-assisted capacity that can infer lung cancer from a limited amount of patient data. To do so, this research starts with a dataset containing 309 patient records, which is augmented and balanced to 540 patient records. This dataset has 16 attributes of symptom and risk-based lung cancer factors. Following preprocessing on the dataset, this study uses a quantum method named quadratic unconstrained binary optimization (QUBO) to select the relevant features of the dataset. Three classifiers, quantum support vector machine (QSVM), quantum k-nearest neighbor (QKNN), and quantum boosting (QBoost), are then trained and tested on the processed dataset to identify lung cancer. These classifiers are examined using different frameworks of circuit and annealing-based quantum techniques. Only 108 balance samples are used during training the classifiers to ensure the learning capability of the quantum method from limited data. To safeguard against overfitting and underfitting due to a lack of data, the remaining 432 samples are evaluated to estimate prediction error. All experiments are done using cross-validation. The final results demonstrate that the QSVM by D-Wave framework achieves the best performance with an accuracy of 94.31% in simulation and 91.90% in a real D-Wave quantum computer. The findings of this work show a path for applying quantum computing to support healthcare decision-making when data availability is often limited.