Leveraging Quantum Kernel Support Vector Machine for breast cancer diagnosis from Digital Breast Tomosynthesis images

Quantum computing, an incredibly emerging technology, provides unique advantages that seamlessly address challenges posed by big data through efficient learning from smaller data sets. Hybrid paradigms like Classical Quantum and Quantum Classical show promise in harnessing this power. We proposed a Quantum Machine Learning (QML) pipeline with QKSVM for diagnosing breast cancer from Digital Breast Tomosynthesis (DBT) images. The prominent objective of this proposed research is to classify the highly complex breast cancer classification from DBT images using appropriate feature extraction and dimensionality reduction in the pipeline integrated with the Quantum Kernel Support Vector Machine (QKSVM) framework of the Quantum Classical paradigm. Six feature extraction methods are employed to extract features from clinical DBT images, followed by pre-processing with Linear Discriminant Analysis (LDA). Classification is then performed using a QKSVM classifier. In the quantum computation phase, different encoding strategies, including Angle, Instantaneous Quantum Polynomial, Quantum Approximate Optimization Algorithm, and custom circuit embeddings, are tested to enhance categorization effectiveness. Comparing classification results with baseline models demonstrates the classifiers’ robustness across various scenarios: changes in feature dimension, use of original and cropped DBT image versions, and application of different quantum kernel feature maps in QKSVM.The proposed method, integrating a 2D feature vector based on LDA with a QKSVM variant utilizing Angle embedding, achieves a noteworthy improvement, with a testing accuracy of around 84.33%. This outperforms other QKSVM variant models. The proposed QKSVM model has been implemented in both the quantum Statevector simulator “SV1” and the real-time QPU “Aspen-M-3” provided by AWS cloud service. The precise classification of clinically cropped images using this QKSVM model confirms its suitability and effectiveness, affirming the robustness of the methodology.