LobePrior - LobePrior segments lung lobes on computed tomography images in the presence of severe abnormalities - automated lung lobe segmentation

Sep 22, 2026

Upload by Adam

LobePrior presents an automated method for lung lobe segmentation on computed tomography in patients with severe pulmonary abnormalities, where manual or even automatic annotation is difficult due to poorly visible fissures. The pipeline uses three stages: coarse low-resolution processing, a high-resolution refinement with specialized AttUNets, and final post-processing. Probabilistic models from label fusion and synthetic lesion generation guide learning in challenging regions. Evaluation on LOLA11 and additional cancer- and COVID-19 datasets shows state-of-the-art Dice scores, including 0.966 on LOCCA and 0.978 on COVID-19 CT.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages17

Reading Time43 min

AccessFree to download

Introduction

Anatomical background and clinical motivation

Challenges in fissure visibility and annotation

LobePrior method overview

Method

Three-stage segmentation pipeline

Probabilistic models and label fusion

Synthetic lesion generation and training

Experiments and Results

Datasets and evaluation metrics

Segmentation performance on LOLA11, LOCCA, and COVID-19 CT

Comparison with competing methods

[www. nature.com/scientificreports](www. nature.com/scientificreports) OPEN LobePrior segments lung lobeson computed tomography images in the presence of severe abnormalities Jean Antonio Ribeiro1􀀍, Diedre Santos do Carmo1, Fabiano Reis2, Ricardo Siufi Magalhães3, Sergio San Juan Dertkigil2, Simone Appenzeller2 & Letícia Rittner1 The development of robust algorithms for lung and lobe segmentation is essential for diagnosing and monitoring pulmonary diseases. Obtaining manual or automatic annotations is challenging, especially in patients with severe abnormalities due to poorly visible lobar fissures. We present LobePrior, an automated lung lobe segmentation method combining deep neural networks and probabilistic models. Segmentation occurs in three stages: a coarse stage processing downsampled images, a highresolution stage where specialized AttUNets segment each lobe, and a final post-processing stage. Probabilistic models derived from label fusion guide the network in regions with severe abnormalities, and synthetic lesion generation provides augmentation during training. Performance was evaluated on LOLA11 and three additional datasets with cancerous nodules or COVID-19 consolidations. LobePrior achieved accurate segmentations compared to manual ground truth, reaching state-of-the-art performance even in challenging cases. On the LOCCA dataset, it obtained a Dice score of 0.966, with similar improvements on a COVID-19 CT dataset (Dice 0.978). Statistically significant improvements over competing methods were observed across all datasets. These results demonstrate that LobePrior effectively integrates anatomical priors and deep learning to provide reliable lobe segmentation in the presence of severe pulmonary abnormalities. Keywords CT Image, Deep Learning, Lung Lobe Segmentation, Medical Image Registration, Prior Information The pulmonary lobes are anatomical subdivisions of the lungs, delineated by pulmonary fissures (Fig. 1-a)) . These structures enable the functional and structural separation of the lungs, facilitating ventilation and gas exchange. They also allow blood circulation through the pulmonary capillaries, where oxygenation occurs. The right lung consists of three lobes: superior, middle, and inferior. The left lung, due to the presence of the heart, is smaller and has two lobes: superior and inferior. Pulmonary lobe segmentation enables quantification of disease involvement and helps characterize conditions such as pulmonary fibrosis, Chronic Obstructive Pulmonary Disease (COPD), lung cancer, and pneumonia, many of which predominantly affect specific lobes or regions of the lungs 1. Cancer, cystic fibrosis, and tuberculosis often affect the upper lobes, whereas interstitial pneumonia and panlobular emphysema frequently involve the lower lobes 1. The location of lung findings can be related to physiological processes. For example, COVID-19 most commonly affects the lower lobes ofthe lungs 2 because this region has greater blood perfusion and ventilation, making it more susceptible to the inflammatory involvement caused by the virus 2. In severe cases, the infection may extend to multiple lobes, resulting in a diffuse viral pneumonia pattern. In addition to the localization of abnormal findings, lung lobe segmentation can enable more precise planning for pulmonary surgeries (such as lobectomies), radiotherapy, and other interventions 3. Manual identification of pulmonary fissures is essential for lobe segmentation, but it is time-consuming and labor-intensive, particularly in high-resolution CT scans 1. Specialists must analyze fissures slice by slice and annotate lobes using tools such as 3D Slicer 4 or ITK-SNAP 5. Because lobar fissures are often partially invisible 1Universidade Estadual de Campinas, School of Electrical and Computer Engineering, Campinas, SP 13083- 970, Brazil. 2Universidade Estadual de Campinas, School of Medical Sciences, Campinas, SP 13083-970, Brazil. 3Faculdade São

FAQs

What makes lung lobe segmentation difficult in severe abnormalities?

Severe abnormalities often make pulmonary fissures poorly visible, which reduces reliability of manual labeling and makes automated annotations challenging.

How does LobePrior perform segmentation?

It uses a three-stage process: low-resolution coarse segmentation, high-resolution refinement with specialized AttUNets, and post-processing of the lung lobes.

Which probabilistic components help LobePrior handle severe cases?

Probabilistic models derived from label fusion guide the network in regions with severe abnormalities, and synthetic lesion generation provides augmentation during training.

Sep 22, 2026

Upload by Adam