Related Experiment Video
Updated: Aug 5, 2026
12:06
Analyzing Mitochondrial Morphology Through Simulation Supervised Learning
Published on: March 3, 2023
Explainable deep learning techniques for microscopic fungi classification using a learnable threshold-based ReLULeaky
Mohsin Hossain1, Mohammad Khairul Islam2, Machbah Uddin3
1Computer Science and Engineering, Uttara University, Dhaka 1230, Bangladesh.
Journal of Pathology Informatics
|July 28, 2026
Summary
This study introduces an explainable AI model for fungal infection diagnosis, achieving high accuracy. The new method improves early detection, crucial for effective treatment and patient outcomes.
Area of Science:
- Medical Mycology
- Artificial Intelligence in Healthcare
- Computer Vision
Background:
- Fungal infections present a growing global health risk, often difficult to diagnose due to resource limitations and complexity.
- Traditional diagnostic methods for fungal infections are time-consuming, expensive, and require specialized expertise, hindering early detection.
- Accurate and rapid diagnosis is critical for timely and effective treatment of fungal infections.
Purpose of the Study:
- To develop an explainable artificial intelligence (AI) model for accurate and efficient classification of fungi.
- To enhance feature representation and classification performance using transfer learning and a novel activation function.
- To improve the speed and accuracy of fungal infection diagnosis through advanced machine learning techniques.
Main Methods:
- A ResNet34 convolutional neural network was fine-tuned using transfer learning.
- A novel learnable threshold-based ReLULeaky activation function was integrated to optimize feature extraction.
- The model's performance was evaluated using metrics including accuracy, F1-score, precision, and area under the curve (AUC).
Main Results:
- The proposed fine-tuned ReLULeaky-ResNet34 model achieved high performance, with an accuracy of 95.39%, F1-score of 96%, and precision of 97%.
- The model demonstrated robust classification capabilities with an AUC score of 99.40%.
- Model interpretability analysis confirmed a focus on biologically relevant morphological features, validating its diagnostic approach.
Conclusions:
- The developed explainable AI model offers a significant advancement in the accurate and rapid classification of fungi.
- The integration of a learnable threshold-based activation function enhances model performance and feature representation.
- This approach holds promise for improving early detection and management of fungal infections, addressing current diagnostic challenges.
