Open-source tool for Airway Segmentation in Computed Tomography using 2.5D Modified EfficientDet: Contribution to the ATM22 Challenge

Abstract

Airway segmentation in computed tomography images can be used to analyze pulmonary diseases, however manual segmentation is labor intensive and relies on expert knowledge. This manuscript details our contribution to MICCAI's 2022 Airway Tree Modelling challenge, a competition of fully automated methods for airway segmentation. We employed a previously developed deep learning architecture based on a modified EfficientDet (MEDSeg), training from scratch for binary segmentation of the airway using the provided annotations. Our method achived 90.72 Dice in internal validation, 95.52 Dice on external validation and 93.49 Dice on the final test phase, while not being specifically designed or tuned for airway segmentation. Open source code and a pip package for predictions with our model and trained weights are in https://github.com/MICLab-Unicamp/medseg.

ElectroNet: An Enhanced Model for Small-Scale Object Detection in Electrical Schematic Diagrams

“…Various machine learning techniques, including Convolutional Neural Networks (CNNs) [5,21,25], Sequential Minimal Optimization (SMO) [3], Support Vector Machine (SVM) classifiers [4,15], K-Nearest Neighbor (KNN) classifiers, and Artificial Neural Networks (ANNs) [24], have been employed to address the challenges inherent in this domain. The evolution of object detection models, from hand-crafted features and classifiers to deep neural networks [30,[41][42][43][44][45]49], has significantly improved computational efficiency and accuracy. Despite these advancements, the field continues to grapple with challenges such as the detection of small objects and variations in drawing styles [50].…”

“…Aldahoul et al [44] used Effi-cientDet for the localization and classification of parasitic eggs in microscopic images, achieving robust performance. Carmo et al [45] used a modified version of EfficientDet for airway segmentation in computed tomography images, achieving high Dice scores. The evolution of object-detection models, from hand-crafted features and classifiers to deep neural networks, has significantly addressed the challenge of computational efficiency and accuracy.…”