MurrAI Quality: development of a deep learning system for the automated evaluation of the quality of sputum samples in clinical microbiology

Identifier:  TFG:9075

Handle: https://hdl.handle.net/20.500.11797/TFG9075

Authors:  Martos i Massó, Laia

Abstract:
Assessment of the quality of sputum samples by Gram staining and a microscope is an essential step in obtaining reliable results in the microbiological diagnosis of respiratory infections of the lower respiratory tract, but its manual interpretation can be subjective and requires time and qualified personnel. In this work, an algorithm called MurrAI Quality has been developed, which is an automated system based on convolutional neural networks capable of detecting and counting squamous epithelial cells and leukocytes in microscopic images to classify the samples according to their quality. The model has been trained with 402 manually annotated sputum microphotographs, indicating the location of leukocytes and epithelial cells in order to provide the model with examples of the patterns to be recognized. Finally, the validity of the system in tasks of segmentation of instances and classification has been evaluated. This study constitutes a proof of concept with potential for implementation in clinical environments, which opens the door to future optimizations.