qml-equilibrium-propagation-medical

name	qml-equilibrium-propagation-medical
description	Quantum Machine Learning with Equilibrium Propagation for medical image analysis. Energy-based training without backpropagation using Variational Quantum Circuits (VQCs) for resource-constrained quantum hardware. Use when: analyzing blood cells, leukemia detection, medical imaging with QML, energy-based quantum training, backprop-free quantum networks, or evaluating QML feasibility on NISQ devices.
metadata	{"arxiv_ids":"1808","published":"2026-01-26","tags":["quantum","machine-learning","medical","equilibrium-propagation","blood-cells","vqc","nisoq"]}

QML with Equilibrium Propagation for Medical Imaging

Description

Feasibility study applying Quantum Machine Learning (QML) with Equilibrium Propagation (EP) — an energy-based, backpropagation-free training method — and Variational Quantum Circuits (VQCs) for acute myeloid leukemia (AML) detection from blood cell images. Demonstrates competitive performance under severe quantum hardware constraints (limited qubits, noise).

Key insight: EP avoids backpropagation by computing gradients through energy differences, making it compatible with quantum circuits where backprop is not natively supported.

Core Methodology

Equilibrium Propagation (EP)

EP computes gradients via energy differences rather than backpropagation:

1. Forward pass: Run input through network to equilibrium state
2. Nudging: Slightly perturb output toward target (±ε)
3. Gradient estimation: ∂E/∂θ ≈ (E_nudged - E_free) / ε
4. Update: Apply gradient to quantum circuit parameters

Variational Quantum Circuits (VQCs)

• Parameterized quantum gates (rotation angles = trainable params)
• Classical-quantum hybrid: classical optimizer updates quantum params
• Measurement outcomes feed into loss function
• Hardware-efficient ansatz for NISQ compatibility

Medical Application Pipeline

1. Input: Blood cell microscopy images
2. Preprocessing: Resize, normalize, encode into quantum states
3. Feature extraction: VQC layer(s) with EP training
4. Classification: Binary (AML vs normal) or multi-class
5. Output: Prediction with confidence

When to Use

• Medical image classification with quantum circuits
• Backpropagation-free quantum training needed
• NISQ hardware constraints (few qubits, high noise)
• Blood cell analysis, leukemia detection
• Evaluating QML feasibility before scaling

Implementation Steps

1. Data preparation: Encode medical images into quantum states (amplitude/angle encoding)
2. VQC design: Choose hardware-efficient ansatz matching device topology
3. EP training: Implement energy-based gradient computation
4. Classical optimizer: Use gradient-free (COBYLA, SPSA) or EP-computed gradients
5. Validation: Compare against classical baselines on same task

Error Handling

Limited Qubit Count

• Use amplitude encoding to maximize information per qubit
• Consider circuit cutting for larger problems
• Feature selection to reduce input dimensionality

Hardware Noise

• Error mitigation: zero-noise extrapolation, readout error correction
• Noise-aware training: include noise model in simulation
• Shallow circuits to reduce decoherence impact

EP Convergence Issues

• Careful ε selection (too small → noisy gradients, too large → biased)
• Multiple nudging directions for variance reduction
• Warm-start from classically pretrained weights

Resources

• Paper: Entity ID 1808 in kg.db
• Related: VQC architecture patterns from quantum-neural-architecture skill
• Related: Medical imaging patterns from quantum-medical-diagnosis skill