Detection of invasive <italic>Saccharomyces cerevisiae</italic> infections based on convolutional neural network

Abstract

<p indent="0mm"><italic>Saccharomyces cerevisiae </italic>(<italic>S</italic>.<italic> cerevisiae</italic>) is common in the baking and brewing industries as a fermenting strain and used as a probiotic for preventing and treating various diarrheal and related diseases. However, the cases of invasive <italic>S</italic>.<italic> cerevisiae</italic> infections due to the ingestion of <italic>S</italic>.<italic> cerevisiae</italic> have been increasing over the last two decades, especially in the elderly, immunocompromised, and critically ill patient populations. Additionally, current clinical methods of diagnosing invasive<italic> S</italic>.<italic> cerevisiae</italic> infections are time-consuming, while the optimal times for the diagnosis and early prevention are easily missed. In this paper, a convolutional neural network (CNN)-based object detection method is proposed for the detection and identification of <italic>S</italic>.<italic> cerevisiae</italic> cells in the blood for invasive <italic>S</italic>.<italic> cerevisiae</italic> disease diagnosis. The method is based on the single shot multibox detector algorithm using the ResNet-50 network as a feature extraction network. The experimental results show that the recognition accuracy of the method can reach 97.70%, with a detection time of <sc>0.31 s.</sc> Moreover, the algorithm recognition accuracy showed a significant advantage over three similar CNN algorithms. The paper verifies that the CNN-based object detection method can be a novel method for the detection and diagnosis of invasive <italic>S</italic>.<italic> cerevisiae</italic> infections, and the advantages of high detection accuracy and short method time can achieve early prevention and timely diagnosis of invasive <italic>S</italic>.<italic> cerevisiae</italic> infections.