Evaluation of Depth-wise Separable Convolution and Channel Attention Mechanism to Bacilli Segmentation

Tuberculosis (TB) is the second leading cause of death from a single infectious agent. In the field of computer vision, much works has been developed to automate the identification of bacilli in sputum smear microscopy field images. In recent years, deep machine learning techniques have been increas...

Full description

Saved in:
Bibliographic Details
Main Authors: Soares, Ronaldo Rodrigues, Monteiro, Felipe Paes, Pinheiro, Gabriel Martins, Serrao, Mikaela Kalline Maciel, Costa Filho, Cicero F. F., Costa, Marly G. F.
Format: Conference Proceeding
Language:English
Subjects:
Attention mechanisms
autoencoder
channel attention mechanism
Computer architecture
Computer vision
Convolution
depth-wise separable convolution
Engineering in medicine and biology
Image segmentation
Microscopy
semantic segmentation
Tuberculosis
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Tuberculosis (TB) is the second leading cause of death from a single infectious agent. In the field of computer vision, much works has been developed to automate the identification of bacilli in sputum smear microscopy field images. In recent years, deep machine learning techniques have been increasingly used. In this work, architectures with different depth, implemented with channel attention mechanism and with depth-wise separable convolution are evaluated to accomplish the task of bacilli segmentation in microscopy images. The best result was obtained with a deeper architecture, implemented with channel attention mechanism and depth-wise separable convolution, with an F1-score of 99.38%.
ISSN:2694-0604
DOI:10.1109/EMBC53108.2024.10781907