Multitasking segmentation of lung and COVID-19 findings in CT scans using modified EfficientDet, UNet and MobileNetV3 models

Carmo, Diedre; Campiotti, Israel; Fantini, Irene; Rodrigues, Lívia; Rittner, Letícia; Lotufo, Roberto de Alencar

doi:10.1117/12.2606118

Cited by 2 publications

(4 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition to quantitative metrics, we also deliver improvements to our graphical (Fig. 3) and command line user interface, providing a open source easy way to predict not only airway segmentations but lung and findings segmentation from our past work [2], including output sheets with volumetric measurements and other statistics.…”

Section: Resultsmentioning

confidence: 99%

“…Our method is named Modified EfficientDet Segmentation (MEDSeg) [2]. MED-Seg is a novel take on EfficientDet [10], a 2D natural image detection network.…”

Section: Methodsmentioning

confidence: 99%

“…Final volumetric segmentations are generated through stacking of 2D (1-channel) predictions, threshold of 0.5, and subsequent largest component extraction post-processing. For this application of MEDSeg, we did not change any parameters previously optimized for COVID-19 findings segmentation [2], and just directly trained once in the provided data. The hidden evaluation of the challenge runners used metrics others than Dice, including Tree length detected rate (TD) and the branches detected rate (BD) [5].…”

Section: Methodsmentioning

confidence: 99%

“…ATM22 advances the availability of annotated airway segmentation data, with an order of magnitude more data than older challenges such as EXACT09 [5]. Our contribution to the challenge consists of employing our Modified EfficientDet Segmentation (MEDSeg) model [2], directly in a 2.5D axial training, only in data from the challenge. No pre-training is involved, and we did not tune the training parameters to this specific dataset with any ablation.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

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

Carmo¹,

Rittner²,

Lotufo³

2022

Preprint

Self Cite

View full text Add to dashboard Cite

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.

show abstract

Section: Resultsmentioning

confidence: 99%

“…Our method is named Modified EfficientDet Segmentation (MEDSeg) [2]. MED-Seg is a novel take on EfficientDet [10], a 2D natural image detection network.…”

Section: Methodsmentioning

confidence: 99%