Automated lung segmentation on chest MRI in children with cystic fibrosis DOI Creative Commons
Friedemann Ringwald, Lena Wucherpfennig, Niclas Hagen

и другие.

Frontiers in Medicine, Год журнала: 2024, Номер 11

Опубликована: Ноя. 12, 2024

Introduction Segmentation of lung structures in medical imaging is crucial for the application automated post-processing steps on diseases like cystic fibrosis (CF). Recently, machine learning methods, particularly neural networks, have demonstrated remarkable improvements, often outperforming conventional segmentation methods. Nonetheless, challenges still remain when attempting to segment various modalities and diseases, especially visual characteristics pathologic findings significantly deviate from healthy tissue. Methods Our study focuses pediatric CF patients [mean age, standard deviation (7.50 ± 4.6)], utilizing deep learning-based methods chest magnetic resonance (MRI). A total 165 standardized annual surveillance MRI scans 84 with were segmented using nnU-Net framework. Patient cases represented a range disease severities ages. The was trained evaluated three sequences (BLADE, VIBE, HASTE), which are highly relevant evaluation induced changes. We utilized 40 training per sequence, tested 15 Sørensen-Dice-Score, Pearson’s correlation coefficient ( r ), questionnaire, slice-based analysis. Results results high level performance across all sequences, only minor differences observed mean Dice coefficient: BLADE (0.96 0.05), VIBE 0.04), HASTE (0.95 0.05). Additionally, quality consistent different severities, patient ages, sizes. Manual identified specific challenges, such as incomplete segmentations near diaphragm dorsal regions. Validation separate, external dataset nine toddlers (2–24 months) generalizability model achieving 0.85 0.03. Discussion conclusion Overall, our demonstrates feasibility effectiveness halves patients, showing promising directions advanced image analysis techniques assist clinical decision-making monitoring progression. Despite these achievements, further improvements needed address enhance generalizability.

Язык: Английский

Automated lung segmentation on chest MRI in children with cystic fibrosis DOI Creative Commons
Friedemann Ringwald, Lena Wucherpfennig, Niclas Hagen

и другие.

Frontiers in Medicine, Год журнала: 2024, Номер 11

Опубликована: Ноя. 12, 2024

Introduction Segmentation of lung structures in medical imaging is crucial for the application automated post-processing steps on diseases like cystic fibrosis (CF). Recently, machine learning methods, particularly neural networks, have demonstrated remarkable improvements, often outperforming conventional segmentation methods. Nonetheless, challenges still remain when attempting to segment various modalities and diseases, especially visual characteristics pathologic findings significantly deviate from healthy tissue. Methods Our study focuses pediatric CF patients [mean age, standard deviation (7.50 ± 4.6)], utilizing deep learning-based methods chest magnetic resonance (MRI). A total 165 standardized annual surveillance MRI scans 84 with were segmented using nnU-Net framework. Patient cases represented a range disease severities ages. The was trained evaluated three sequences (BLADE, VIBE, HASTE), which are highly relevant evaluation induced changes. We utilized 40 training per sequence, tested 15 Sørensen-Dice-Score, Pearson’s correlation coefficient ( r ), questionnaire, slice-based analysis. Results results high level performance across all sequences, only minor differences observed mean Dice coefficient: BLADE (0.96 0.05), VIBE 0.04), HASTE (0.95 0.05). Additionally, quality consistent different severities, patient ages, sizes. Manual identified specific challenges, such as incomplete segmentations near diaphragm dorsal regions. Validation separate, external dataset nine toddlers (2–24 months) generalizability model achieving 0.85 0.03. Discussion conclusion Overall, our demonstrates feasibility effectiveness halves patients, showing promising directions advanced image analysis techniques assist clinical decision-making monitoring progression. Despite these achievements, further improvements needed address enhance generalizability.

Язык: Английский

Процитировано

0