By name: qcnn-parallel-feature-fusion-medical description: "Parallel multi-circuit quantum feature fusion methodology for medical image classification. Use when: (1) building hybrid quantum-classical CNN architectures for biomedical image classification, (2) comparing quantum vs classical models with statistical rigor (Wilcoxon signed-rank test, Cohen's d effect size), (3) designing parallel quantum encoding circuits (amplitude + angle encoding simultaneously), (4) parameter-matched fairness evaluation for QML vs classical baselines. Covers QCNN architecture pattern, dual-encoding VQC fusion, and NISQ-era validation framework." license: Complete terms in LICENSE.txt metadata: arxiv_id: "2512.02066" published: "2025-11-29" authors: "Ece Yurtseven" tags: [quantum, medical, qcnn, feature-fusion, statistical-validation, breast-cancer, breastmnist]
Parallel Multi-Circuit Quantum Feature Fusion for Medical Imaging
Core Methodology
Hybrid QCNN architecture that runs two distinct quantum circuits in parallel:
- Amplitude-encoding VQC — encodes classical features as quantum state amplitudes
- Angle-encoding VQC with circular entanglement — encodes features as rotation angles with ring topology entanglement
Both operate on 4 qubits. Their quantum feature embeddings are fused with classical convolutional features into a joint feature space, then processed by a fully connected classifier.
Key Innovation: Statistical Validation Framework
Establishes rigorous statistical comparison between hybrid quantum and classical models:
- Parameter matching — quantum and classical models have matched parameter counts to isolate quantum contribution
- Multiple independent runs — 5 independent training runs for statistical significance
- Wilcoxon signed-rank test — non-parametric test (p = 0.03125) confirms significance
- Cohen's d effect size — large effect (d = 2.14) confirms practical significance, not just statistical
Architecture Pattern
Input Image → Classical Conv Layers → Feature Maps
↓
┌────────────────────┼────────────────────┐
↓ ↓ ↓
Amplitude-Enc VQC Angle-Enc VQC Classical Features
(4 qubits, 4 qubits) (circular entang.) (conv outputs)
↓ ↓ ↓
Quantum Emb 1 Quantum Emb 2 Classical Vectors
└────────────────────┼────────────────────┘
↓
Joint Feature Space
↓
Fully Connected Classifier
↓
Binary Classification
When to Use This Skill
- Medical image classification tasks with limited data (BreastMNIST, ChestX-ray, etc.)
- NISQ-era experiments with 4-8 qubits
- Need statistically validated quantum advantage claims
- Comparing QML models against classical baselines with fair parameter budgets
- Building hybrid architectures that fuse multiple encoding strategies
Implementation Steps
Step 1: Classical Feature Extraction
# Standard CNN backbone (parameter-matched to baseline)
class ConvBackbone(nn.Module):
def __init__(self, base_params=50000):
super().__init__()
self.conv1 = nn.Conv2d(1, 16, 3, padding=1)
self.conv2 = nn.Conv2d(16, 32, 3, padding=1)
self.pool = nn.AdaptiveAvgPool2d((4, 4))
# Parameter count must match classical baseline
Step 2: Parallel Quantum Circuits
# Amplitude encoding VQC
def amplitude_encoding_vqc(features, n_qubits=4):
# Encode classical features as quantum amplitudes
# Apply variational layers
return quantum_state_measurements
# Angle encoding VQC with circular entanglement
def angle_encoding_vqc(features, n_qubits=4):
# Encode features as rotation angles (RY gates)
# Apply circular (ring) entanglement pattern
# Apply variational layers
return quantum_state_measurements
Step 3: Feature Fusion & Classification
def forward(x):
classical_features = conv_backbone(x)
quantum_amp = amplitude_encoding_vqc(classical_features)
quantum_angle = angle_encoding_vqc(classical_features)
# Fuse all embeddings
joint = torch.cat([classical_features.flatten(),
quantum_amp, quantum_angle], dim=1)
return classifier(joint)
Step 4: Statistical Validation
from scipy import stats
import numpy as np
# After 5+ independent runs with same seed splits
quantum_accs = [acc1, acc2, acc3, acc4, acc5]
classical_accs = [acc1, acc2, acc3, acc4, acc5]
# Wilcoxon signed-rank test (paired, one-sided)
stat, p_value = stats.wilcoxon(quantum_accs, classical_accs,
alternative='greater')
# Cohen's d effect size
d = (np.mean(quantum_accs) - np.mean(classical_accs)) / pooled_std
# Report: p < 0.05 and d > 0.8 (large) = statistically and practically significant
Error Handling
Quantum Advantage Not Significant
- Check parameter matching — quantum model must not have more trainable parameters
- Increase number of independent runs (5 minimum, 10 recommended)
- Try different encoding strategies (this paper shows angle+amplitude fusion works)
- Consider dataset size — small datasets may not show advantage
NISQ Hardware Limitations
- 4-qubit circuits are NISQ-friendly; scale cautiously
- Use noise mitigation if deploying on real hardware
- Simulation first, then validate on hardware
Parameter Matching Pitfall
- Count ALL trainable parameters including quantum gate angles
- Classical baseline should have same total parameter budget
- Document the matching methodology for reproducibility
Verified Results
- Dataset: BreastMNIST (binary benign/malignant classification)
- Hybrid QCNN: Statistically significant improvement over parameter-matched CNN
- p = 0.03125 (Wilcoxon, one-sided)
- Cohen's d = 2.14 (large effect size)
- Published: QCNC 2026
Activation Keywords
- qcnn parallel feature fusion
- quantum feature fusion medical
- statistical validation quantum ml
- wilcoxon quantum advantage
- cohen d quantum classification
- amplitude angle encoding fusion
- breastmnist quantum classification
- parameter matched quantum baseline
- parallel vqc medical imaging