Trypanosoma cruzi Detection using LSTM Convolutional Autoencoder

Abstract

The presence of Trypanosoma cruzi (T. cruzi) parasites in blood samples is proof of the medical diagnosis of Chagas disease. Since the motion of these microorganisms is conspicuous in optical microscopy videos, we propose a spatio-temporal autoencoder for anomaly detection caused by parasite motility. This approach includes a spatial feature extractor and a temporal sequencer ConvLSTM for learning the temporal evolution of the spatial features. We trained the autoencoder with no parasites videos to learn the normal pattern and measured the regularity score in test videos with parasites. Our results showed that an LSTM-based autoencoder may identify T. cruzi anomalous motion, being a promising method for detecting parasites in microscopy videos.