Authors

Cairo University - Biomedical Engineering Department

• Ahmed Loay ElSayed
• Lyan Ahmed Mohsen
• Sarah Sameh Mohamed
• Mariam Sherif Mohamed
• Alhussien Ayman Hanafy

ACO-RFPDE Biopsy Image Denoising

A novel framework for denoising Poisson-corrupted microscopic biopsy images using Ant Colony Optimization and Fourth-Order Partial Differential Equations.

Overview

Removes Poisson noise from lung cancer biopsy images while preserving diagnostic features. Combines ACO optimization with PDE-based filtering and BM3D enhancement.

Key Features:

• Poisson noise removal with edge preservation
• Automated parameter tuning via ACO
• 95%+ accuracy across noise levels 1-8
• Works on lung and colon cancer biopsies

🚀 Installation Prerequisites :

• Python 3.7+
• NumPy
• SciPy
• OpenCV
• Matplotlib
• scikit-image

Quick Usage

from aco_rfpde import BiopsyDenoiser
import cv2

# Load and denoise image
denoiser = BiopsyDenoiser()
noisy_image = cv2.imread('noisy_biopsy.jpg')
denoised = denoiser.denoise(noisy_image, noise_level=4)
enhanced = denoiser.enhance_bm3d(denoised)

cv2.imwrite('result.jpg', enhanced)

Results

Visual Results

Benign Lung Tissues

Performance Metrics

Benign Lung Tissues

Noise Level	Stage	PSNR	SSIM	MSE	NAE	CC	UQI
4	Noisy	13.348	0.084	0.046	0.239	0.506	0.440
4	PDE Denoised	21.534	0.306	0.007	0.090	0.847	0.837
4	BM3D Enhanced	31.527	0.901	0.001	0.030	0.988	0.987
6	Noisy	5.296	0.014	0.296	0.630	0.166	0.092
6	PDE Denoised	21.508	0.334	0.007	0.097	0.869	0.859
6	BM3D Enhanced	27.922	0.905	0.002	0.049	0.989	0.988

Lung Adenocarcinomas

Noise Level	Stage	PSNR	SSIM	MSE	NAE	CC	UQI
4	Noisy	13.387	0.094	0.046	0.287	0.439	0.355
4	PDE Denoised	21.862	0.365	0.007	0.107	0.799	0.779
4	BM3D Enhanced	31.764	0.859	0.001	0.034	0.975	0.972
6	Noisy	5.663	0.016	0.274	0.782	0.145	0.070
6	PDE Denoised	22.146	0.403	0.006	0.119	0.832	0.816
6	BM3D Enhanced	29.077	0.865	0.001	0.049	0.977	0.976

Lung Squamous Cell Carcinomas

Noise Level	Stage	PSNR	SSIM	MSE	NAE	CC	UQI
4	Noisy	13.444	0.062	0.045	0.299	0.397	0.302
4	PDE Denoised	21.710	0.271	0.006	0.115	0.755	0.726
4	BM3D Enhanced	32.926	0.874	0.001	0.031	0.977	0.974
6	Noisy	5.742	0.010	0.269	0.820	0.129	0.055
6	PDE Denoised	21.185	0.270	0.008	0.146	0.765	0.735
6	BM3D Enhanced	28.773	0.871	0.001	0.055	0.978	0.975

Performance Graphs

Line plots showing PSNR, SSIM, MSE, NAE, CC, and UQI improvements across processing stages

Key Metrics Explanation

• PSNR: Peak Signal-to-Noise Ratio (higher = better)
• SSIM: Structural Similarity Index (closer to 1 = better)
• MSE: Mean Squared Error (lower = better)
• NAE: Normalized Absolute Error (lower = better)
• CC: Correlation Coefficient (closer to 1 = better)
• UQI: Universal Quality Index (closer to 1 = better)

Dataset

Uses LC25000 dataset with lung and colon cancer histopathological images.

Citation

@article{elsayed2024denoising,
  title={Denoising Of Poisson-Corrupted Biopsy Images},
  author={ElSayed, Ahmed Loay and Mohsen, Lyan Ahmed and Mohamed, Sarah Sameh and Mohamed, Mariam Sherif and Hanafy, Alhussien Ayman},
  journal={Cairo University},
  year={2024}
}