Related Experiment Video
Updated: Jul 9, 2026
11:36
Cutaneous Leishmaniasis in the Dorsal Skin of Hamsters: a Useful Model for the Screening of Antileishmanial Drugs
Published on: April 21, 2012
22.7K
A lightweight deep learning framework for reliable microscopy-based diagnosis of cutaneous leishmaniasis
Nisreen Osman E Ahmed1,2, Samuel Mwalili3, Murtada K Elbashir4,5
1Department of Mathematics, Institute for Science, Technology and Innovation, Pan African University, Nairobi, Kenya.
Plos One
|March 17, 2026
Summary
A new deep learning framework automates cutaneous leishmaniasis (CL) diagnosis from microscopy images. Isotonic calibration significantly improves diagnostic probability reliability, enhancing this tool for One Health applications.
Area of Science:
- Medical Diagnostics
- Artificial Intelligence
- Parasitology
Background:
- Cutaneous leishmaniasis (CL) diagnosis relies on microscopy, which is time-consuming and operator-dependent.
- There is a need for automated, reliable diagnostic tools for CL.
Purpose of the Study:
- To develop a lightweight, calibration-aware deep learning framework for automated amastigote detection and slide-level diagnosis of CL.
- To evaluate the impact of post-hoc calibration techniques on diagnostic probability reliability.
Main Methods:
- A U-Net architecture with a MobileNetV2 encoder was used for pixel-level parasite segmentation.
- Weakly supervised pseudo-labeling and probability pooling were employed.
- Isotonic regression, Platt scaling, and temperature scaling were used for post-hoc calibration.
Main Results:
- The framework achieved high segmentation performance (Dice: 0.901, IoU: 0.820).
- Strong slide-level discrimination was observed (AUROC: 0.978).
- Isotonic calibration significantly improved probability reliability (Brier score: 0.030, ECE: 0.023) without compromising discrimination.
Conclusions:
- The developed deep learning framework offers a robust and reliable method for automated CL diagnosis.
- Isotonic calibration enhances the interpretability and trustworthiness of AI-driven diagnostic probabilities.
- This framework supports scalable microscopy-based screening and One Health initiatives.
