Automated classification of Echinococcus multilocularis protoscolex viability using machine learning for high-throughput drug screening.

Abstract

Alveolar echinococcosis, caused by the parasite Echinococcus multilocularis, is a potentially lethal disease that requires the development of new medical treatments. Current drug screening methods depend on manual microscopy to count stained and unstained protoscoleces to determine viability, a process that is often slow and subject to human error. This study developed machine learning models to automate the classification of protoscolex survival to accelerate drug discovery. A total of 111 images containing 1,302 protoscoleces stained with eosin or methylene blue were analyzed. Histogram-based image features were extracted and subjected to preprocessing, dimensionality reduction, and feature selection. Six classification algorithms were trained using five-fold cross-validation and evaluated on internal test sets and an external validation set via receiver operating characteristic curve analysis. Results indicated that several models were highly effective, with internal test scores ranging from 0.8905 to 0.9467 and external validation scores between 0.7772 and 0.8214. Models based on support vector machine, logistic regression, and linear discriminant analysis demonstrated promising discriminatory performance across both internal and external datasets. The study concludes that machine learning can accurately and objectively classify the viability of protoscoleces based on stained images. This automated approach offers significant potential for standardizing high-throughput drug screening and facilitating the discovery of new therapeutics against this neglected zoonotic disease.

Abstract as supplied by the indexing source. © the authors / publisher.