Quantum Machine Learning Achieves 86.4% Accuracy Detecting Leukemia with 50 Samples

Detecting Acute Myeloid Leukemia (AML) from microscopic images of blood cells presents a significant challenge, but new research indicates quantum machine learning could offer a viable solution. A. Bano and L. Liebovitch, alongside their colleagues, demonstrate the potential of algorithms like Equilibrium Propagation and Variational Quantum Circuits to accurately identify AML, even with limited data and computational resources. Their study, utilising the AML-Cytomorphology dataset, achieves surprisingly competitive results , reaching up to 86.4% accuracy with Equilibrium Propagation , despite operating with reduced image resolution and fewer training samples than conventional methods. This work establishes crucial baselines for near-term quantum computing applications in healthcare, suggesting that quantum machine learning is a feasible approach for medical image analysis even in the NISQ era.

This study utilized the AML-Cytomorphology dataset, a collection of 18,365 expert-annotated images, and focused on establishing reproducible baselines for QML in healthcare, validating its potential within the current Noisy Intermediate-Scale Quantum (NISQ) era. The team achieved this by training models on limited subsets of the dataset, ranging from 50 to 250 samples per class, and deliberately introducing limitations such as reduced image resolution (64×64 pixels) and engineered features (20D) to simulate realistic computational scenarios.

Experiments show that these quantum methods achieved performance levels only 12-15% below those of classical Convolutional Neural Networks (CNNs), a remarkable result given the imposed limitations. Specifically, the EP algorithm, which crucially avoids backpropagation, a process incompatible with quantum systems due to state-collapsing measurements, reached an accuracy of 86.4%, just 12% below the performance of the CNN. The 4-qubit VQC, utilizing a ZZFeatureMap encoding and a shallow RealAmplitudes ansatz, attained 83.0% accuracy, exhibiting consistent data efficiency; it maintained a stable 83% performance even with only 50 samples per class, whereas the CNN required 250 samples, five times more data, to achieve 98% accuracy. This data efficiency is particularly noteworthy, as expert annotations in medical domains are often costly and time-consuming to obtain.
The research establishes a reproducible EP pipeline for AML detection, leveraging engineered image features and training without backpropagation, and presents a benchmark across varying dataset scales to assess QML performance and runtime. By performing these simulations on a standard laptop computer using the IBM Qiskit quantum simulator, the scientists demonstrated that even exploratory studies can yield valuable insights into the potential of these novel QML algorithms. This work addresses key challenges in the field, including scaling beyond toy datasets, handling real-world data variability, and demonstrating practical advantages over classical methods, paving the way for future advancements in quantum-enhanced medical diagnostics. To facilitate this, they utilised the AML-Cytomorphology dataset, a collection of 18,365 expert-annotated images, and strategically selected subsets ranging from 50 to 250 samples per class for training and testing. Experiments employed a rigorous methodology involving image pre-processing and feature engineering to reduce computational demands, images were downscaled to 64×64 pixels, and a 20-dimensional feature space was engineered from the original data.

The team then implemented EP, an energy-based learning method that circumvents the need for backpropagation, a process incompatible with quantum systems due to state-collapsing measurements. EP was trained to classify AML cells without relying on gradients, instead leveraging equilibrium states derived from free and nudged dynamics, a crucial innovation for potential quantum hardware implementation. Furthermore, the study pioneered a 4-qubit VQC classifier, utilising a ZZFeatureMap for encoding classical data into quantum states and a shallow RealAmplitudes ansatz for circuit parameter optimisation. This VQC was designed to maintain consistent performance even with limited data, demonstrating its data efficiency, achieving 83% accuracy with only 50 samples per class, while a classical CNN required 250 samples to reach 98% accuracy.

All quantum simulations were performed using the IBM Qiskit platform on a standard laptop computer, establishing reproducible baselines for QML in healthcare. The results revealed that EP achieved 86.4% accuracy, only 12% below the performance of classical CNNs, while the VQC attained 83.0% accuracy, demonstrating competitive performance despite operating under significant constraints. This work validates the potential of QML for near-term applications in healthcare, offering a pathway towards leveraging quantum resources for improved medical diagnostics.

AML Detection via Quantum Machine Learning

Scientists achieved competitive performance with quantum machine learning (QML) algorithms on real-world medical imaging, even under significant constraints. Key to this work was the use of the AML-Cytomorphology dataset, containing 18,365 expert-annotated images, to rigorously test the feasibility of QML in a healthcare setting. Experiments revealed that quantum methods achieved performance levels only 12-15% below those of classical Convolutional Neural Networks (CNNs), despite operating with reduced image resolution (64×64 pixels) and utilising engineered features limited to 20 dimensions.

The EP algorithm, crucially, reached 86.4% accuracy without employing backpropagation, a technique incompatible with quantum systems due to state-collapsing measurements, representing a 12% performance gap compared to the CNN benchmark. Data shows the 4-qubit VQC attained 83.0% accuracy, demonstrating consistent data efficiency; it maintained a stable 83% performance level using only 50 samples per class. Tests prove that the VQC requires five times less data than the CNN, which needs 250 samples to achieve 98% accuracy, highlighting a potential advantage in scenarios where labelled data is scarce. The team measured performance across varying dataset scales, ranging from 50 to 250 samples per class, to establish reproducible baselines for QML in healthcare and validate its feasibility within the current Noisy Intermediate-Scale Quantum (NISQ) era.

This breakthrough delivers a functional EP pipeline for AML detection, trained without backpropagation, and a 4-qubit VQC classifier employing ZZFeatureMap encoding and a shallow RealAmplitudes ansatz. Measurements confirm that this study systematically compares quantum-inspired (EP) and pure quantum (VQC) approaches under identical experimental conditions, quantifying data efficiency, a critical factor for medical applications where expert annotations are costly and time-consuming. The research was performed using a laptop computer and the IBM Qiskit quantum simulator, demonstrating that even exploratory simulations can yield valuable insights into the performance of these novel QML algorithms. These results establish a foundation for future work exploring more complex QML architectures and their application to a wider range of medical imaging challenges.

AML Detection via Quantum Machine Learning

Scientists have demonstrated the feasibility of applying quantum machine learning (QML) algorithms to real-world medical imaging tasks. Key results indicate that, despite operating with limited data (50-250 samples per class) and reduced image resolution, quantum methods achieved performance within 12-15% of classical Convolutional Neural Networks (CNNs). . Notably, EP reached 86.4% accuracy without utilising backpropagation, a technique incompatible with quantum systems, while the 4-qubit VQC attained 83.0% accuracy and maintained consistent performance even with only 50 samples per class, a data efficiency exceeding that of CNNs which required 250 samples to achieve peak performance.

This suggests a potential advantage for QML in scenarios where labelled data is scarce, such as rare disease research. The shallow circuit design employed is compatible with current Noisy Intermediate-Scale Quantum (NISQ) hardware, facilitating near-term validation. . The authors acknowledge limitations including the current superiority of classical CNNs in absolute accuracy for this specific classification task.  Future research will focus on shot-based sampling experiments, error mitigation strategies like zero-noise extrapolation, analysis of barren plateaus during training, and the development of hybrid quantum-classical architectures.  Furthermore, validation across diverse patient populations is planned.  These findings establish reproducible baselines for QML in healthcare, indicating that quantum methods may offer a viable alternative when data availability, rather than model capacity, presents the primary constraint.