🧬 Explainable Noise-Resilient Quantum Neural Networks for Breast Cancer Detection

Hybrid Classical-Quantum Deep Learning System for Mammogram Classification

---

🚀 Live Demo

🌐 https://explainable-noise-resilient-quantum.onrender.com

---

🧠 Research Goal

Most deep learning systems for cancer detection behave as black boxes — they output predictions but do not explain how the decision was made.

This research explores whether quantum circuits can extract interpretable diagnostic features from mammogram images.

The system integrates a classical CNN feature extractor with quantum variational circuits and studies their performance under realistic quantum noise conditions.

---

🏗 System Architecture

Pipeline:

Mammogram Image
        ↓
DenseNet121 Feature Extractor
        ↓
Feature Compression Layer
        ↓
Quantum Circuit Layer
        ↓
Quantum Expectation Values ⟨Z⟩
        ↓
Final Classifier
        ↓
Benign / Malignant

---

🔬 Quantum Circuit Configurations

Three architectures are tested to study quantum contribution.

1️⃣ Single Qubit Rotation Model

• Qubits: 1 • Gates: RX, RY, RZ • Entanglement: None

Purpose: Evaluate minimal quantum representation.

---

2️⃣ Entanglement-Only Model

• Qubits: 4 • Gates: Hadamard + PhaseShift + CNOT • No rotational gates

Purpose: Measure diagnostic signal from pure quantum correlation.

---

3️⃣ Full Variational Quantum Circuit

• Qubits: 4 • Gates: RX, RY, RZ + CNOT • 2 variational layers

Purpose: Maximum quantum expressibility.

---

🧪 Noise Robustness Study

The model is evaluated under four quantum noise channels.

Noise Type	Description
Depolarizing	Random Pauli errors
Bit Flip	Qubit state flip
Phase Flip	Phase inversion
Amplitude Damping	Energy decay

Noise probabilities tested:

p = 0.01
p = 0.05
p = 0.10

This ensures the model is viable on NISQ-era quantum hardware.

---

📊 Model Output Example

{
 "results": {
  "single_qubit": {"prediction": "Benign", "confidence": 94.4},
  "entanglement": {"prediction": "Benign", "confidence": 93.1},
  "full_variational": {"prediction": "Benign", "confidence": 84.6},
  "ensemble": {"prediction": "Benign", "confidence": 90.7}
 }
}

---

🛠 Tech Stack

Component	Technology
Quantum Framework	PennyLane
Deep Learning	PyTorch
CNN Backbone	DenseNet121
API Backend	FastAPI
Deployment	Render
Containerization	Docker

---

📂 Project Structure

quantum-api/

app.py
quantum_models.py
requirements.txt
Dockerfile

models/
 ├── single_qubit_best.pth
 ├── entanglement_best.pth
 └── full_variational_best.pth

static/
 └── index.html

assets/
 ├── demo.png
 ├── architecture.png
 └── prediction.png

---

⚙️ Run Locally

git clone https://github.com/YOUR_USERNAME/quantum-breast-cancer-api.git

cd quantum-breast-cancer-api

pip install -r requirements.txt

uvicorn app:app --reload

Open:

http://localhost:8000

---

🏋 Training Configuration

Parameter	Value
Dataset	CBIS-DDSM
Image Size	224 × 224
Epochs	20
Optimizer	Adam
Loss	BCEWithLogits
Training Platform	Kaggle GPU

---

⚠ Disclaimer

This system is developed only for academic research purposes and should not be used for clinical diagnosis.

---

👨‍💻 Author

• P V Akhila , MSc Data Science Dissertation Hybrid Classical-Quantum Neural Networks for Medical Imaging