Enhancing leptospirosis screening using a deep convolutional neural network with microscopic agglutination test...

Area of Science:

• Medical Diagnostics
• Bioinformatics
• Computational Biology

Background:

• Leptospirosis is a significant global public health concern, particularly endemic in Malaysia with seasonal peaks.
• The microscopic agglutination test (MAT) is the standard for leptospirosis confirmation but is labor-intensive and subjective.
• Current diagnostic methods for leptospirosis present challenges in terms of efficiency and accuracy.

Purpose of the Study:

• To develop a semiautomated workflow for Leptospira screening.
• To integrate a Deep Convolutional Neural Network (DCNN) with the conventional MAT.
• To improve the speed and reduce inaccuracies in leptospirosis diagnosis.

Main Methods:

• Development of a hybrid diagnostic workflow combining TensorFlow and Keras-based DCNN with MAT.
• Training the DCNN model on a dataset of 442 positive and 442 negative MAT images of Malaysian Leptospira serovars.
• Hyperparameter tuning of the DCNN model, including convolutional layers, filters, kernel sizes, dense layer units, activation functions, and learning rate.

Main Results:

• The developed model achieved a Precision score of 0.8125, a Recall of 0.9286, and an F1-Score of 0.8667 when compared to verified patient MAT results.
• The semiautomated workflow demonstrated significant improvements in speed and accuracy over conventional methods.
• The model's performance indicates its potential for practical application in Leptospira diagnosis.

Conclusions:

• The hybrid diagnostic approach offers a significant advancement in managing leptospirosis in Malaysia.
• The developed DCNN-integrated MAT workflow is practical, adaptable, and suitable for other laboratories diagnosing Leptospira.
• Further scaling of the dataset can enhance model accuracy and adaptability for use in other endemic regions.