Brain Tumor Classification using Hybrid QCNN with ResNet

Hybrid quantum–classical model combining a small Quantum FiLM modulation (PennyLane) applied before a ResNet-50 backbone. The model classifies MRI images into multiple brain tumour categories (glioma, meningioma, pituitary) plus no_tumor. This repo includes training, inference (Gradio), and baseline comparison scripts.

---

Motivation

Brain tumour diagnosis using MRI requires expert radiologists and can be slow or inconsistent.
This project explores a Hybrid Quantum–Classical approach to improve feature extraction and
decision boundaries using a Quantum FiLM modulation layer applied before a ResNet-50 backbone.

The goal is to evaluate whether quantum-inspired methods can improve classification stability
and generalization for medical imaging tasks.

---

Features

• Quantum FiLM modulation (per-image scale & shift) computed by a 4-qubit PennyLane circuit
• ResNet-50 backbone adapted for grayscale medical images
• End-to-end training & validation pipeline with checkpoint saving
• Gradio-based image upload UI for quick demo / inference
• Comparison script to evaluate a standard ResNet baseline
• Designed for reproducible experiments and easy extension

---

Tech Stack

• Python 3.8+
• PyTorch & torchvision (modeling & training)
• PennyLane (quantum circuit + TorchLayer)
• NumPy, PIL (data handling)
• scikit-learn (metrics)
• Gradio (prediction UI)

Utilities

• torch.optim optimizers & LR schedulers
• torchvision.transforms for augmentation & normalization
• DataLoader, random_split for reproducible train/val splits

---

Key Performance Highlights

• 99% overall validation accuracy
• High F1-scores across all tumour types (0.97–0.99)
• Perfect recall (1.00) for pituitary tumour
• Quantum FiLM improves global feature modulation
• ResNet-50 extracts rich hierarchical features
• Strong generalization on 4-class MRI dataset

---

Dataset

Kaggle dataset by the author:
https://www.kaggle.com/datasets/skarthik112/karthik-braintypesdata-mri

Dataset Folder Structure

brain_Tumor_Types/
│
├── glioma/
├── meningioma/
├── pituitary/
└── no_tumor/

---

Sample Results

Glioma Detected

Meningioma Detected

Pituitary Tumour Detected

No Tumour Detected

---

Repository Structure

Brain-Tumor-QCNN-ResNet/
│
├── train_hybrid.py              
├── predict_gradio.py            
├── compare_models.py            
├── requirements.txt             
├── README.md                    
├── .gitignore                   
│
└── assets/                      
    ├── bannerdl.jpg
    ├── giloma result.png
    ├── meningioma result.png
    ├── pituitary result.png
    └── no tumor result.png

---

Classification Report

Class	Precision	Recall	F1-Score	Support
Glioma	0.99	0.97	0.98	277
Meningioma	0.97	0.96	0.97	255
No Tumor	0.98	0.99	0.99	322
Pituitary	0.98	1.00	0.99	289
Accuracy	—	—	0.99	1143
Macro Avg	0.98	0.98	0.98	1143
Weighted Avg	0.98	0.98	0.98	1143

---

Author

S. Karthik
Developer & Research Student
Brain Tumor Classification using Hybrid QCNN with ResNet (2025)

---

Installation

Follow the steps below to set up the environment and run the Hybrid QCNN + ResNet model.

Download or Clone the Repo

git clone https://github.com/Karthik7661/Brain-Tumor-QCNN-ResNet.git
cd Brain-Tumor-QCNN-ResNet
#2 — Install Required Dependencies
pip install -r requirements.txt
#3 — Train the Hybrid QCNN + ResNet Model
python train_hybrid.py
#4 — Run the Gradio Prediction Application
python predict_gradio.py