Related Experiment Video
Updated: Apr 30, 2026
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
2.0K
LExNet: A bio-inspired lightweight ensemble model for breast cancer classification using hybrid autoencoder and swarm
Roseline Oluwaseun Ogundokun1,2,3, Pius Adewale Owolawi1, Etienne van Wyk1
1Department of Computer Systems Engineering, Tshwane University of Technology (TUT), Pretoria, South Africa.
Digital Health
|April 29, 2026
Summary
LExNet, a novel deep learning model, enhances breast cancer diagnosis accuracy to 98.3% using an ensemble of lightweight networks and autoencoder features. This efficient, interpretable solution aids real-time diagnostics, especially in resource-limited settings.
Area of Science:
- * Medical image analysis
- * Computational pathology
- * Artificial intelligence in oncology
Background:
- * Breast cancer diagnosis relies on histopathology, which is labor-intensive and prone to variability.
- * Accurate and timely diagnosis is crucial for effective breast cancer treatment.
- * Existing methods face challenges in efficiency and consistency.
Purpose of the Study:
- * To introduce LExNet, a bio-inspired lightweight ensemble model for breast cancer diagnosis.
- * To improve diagnostic accuracy and computational efficiency in histopathological image analysis.
- * To provide an interpretable and robust solution for real-time breast cancer detection.
Main Methods:
- * Development of LExNet, an ensemble of lightweight Convolutional Neural Network (CNN) architectures (MobileNetV3, EfficientNet-B0, ShuffleNet-V2, SqueezeNet).
- * Integration of hybrid autoencoder-based feature extraction for noise and dimensionality reduction.
- * Application of Particle Swarm Optimization (PSO) for efficient hyperparameter tuning.
- * Utilization of Grad-CAM for model interpretability analysis.
Main Results:
- * LExNet achieved a validation accuracy of approximately 98.3% on the ICIAR 2018 BACH dataset.
- * Demonstrated a 40-50% reduction in computational demands compared to traditional models.
- * Reduced overfitting risk by over 30% through autoencoder feature extraction.
- * Decreased manual hyperparameter tuning time by up to 70% using PSO.
- * Grad-CAM analysis confirmed alignment of predictions with expert pathologist insights.
Conclusions:
- * LExNet offers a highly accurate and computationally efficient solution for breast cancer diagnosis.
- * The model's interpretability enhances clinical relevance and trust.
- * LExNet is particularly suitable for resource-constrained clinical settings, enabling real-time diagnostics.