Author ORCID Identifier

Year of Publication


Degree Name

Master of Science in Biomedical Engineering

Document Type

Master's Thesis




Biomedical Engineering

First Advisor

Dr. Abhijit Patwardhan


Approximately a third of the population worldwide is chronically infected by the parasite Toxoplasma gondii. During chronic infection the parasite resides within tissue cysts as the poorly understood bradyzoite form. Observation of these bradyzoites via microscopic imaging within tissue cyst purified from infected mouse brains has shown metabolic activity with heterogeneous replication potential. With fluorescence microscopy imaging targeting the parasite’s actively respiring mitochondria, the parasite’s metabolic state can be further investigated as the morphology of mitochondria can be associated with specific physiologic states. However, manually classifying mitochondrial morphologies from these images can be tedious and prone to error as the bradyzoites within cysts can number into thousands. Towards this end, computer based tools were developed to facilitate and automate the identification and classification of the parasite’s mitochondrial morphologies to aid the study of bradyzoite biology. The developed image processing based program assists the manual classification of mitochondrial morphologies by trained operators while collecting features and statistics from the manual classification of shapes. The manual classifications, obtained from a subset of images was used as the gold standard to train a multivariate logistic regression algorithm. Results from the machine learning based automatic classification, trained on a total of 1,138 discrete mitochondrial objects from 5 images of individual cysts, and tested on 5 different images of individual cysts showed that from a total of 927 discrete mitochondrial objects that were identified, an average overall accuracy of 82% was obtained in predicting the mitochondrial morphology as belonging to one of the five predefined classes of Blobs, Tadpoles, Donuts, Arcs, or Other. The Lasso morphology, typically associated with actively growing parasites was not observed with significant frequency. The majority of detected objects, 55%, of the objects from the 5 images used in the training set were identified as belonging to the Blob morphology which is presumptively associated with low bradyzoite metabolic activity. There was a lower incidence of Tadpole and Arc morphologies which are associated with more active parasites. The performance of the trained machine learning algorithm resulted in a high degree of confidence in the prediction of Blobs with an average F score of 0.91 and Arcs with an average F score of 0.73 which added together make up a majority of morphological classes present in most bradyzoites, 85% of the objects detected in the training set. These results indicate that the developed approach can advance the investigation of bradyzoite biology by allowing for a higher throughput of images analyzed and thus potentially assist in the evaluation of the efficacy of potential drug treatments.

Digital Object Identifier (DOI)

Funding Information

This study was supported by the National Institutes of Health grant (no.: R01AI145335) from 2019 to 2021.