eLife · 2020
ODELAM, rapid sequence-independent detection of drug resistance in isolates of Mycobacterium tuberculosis.
СтатьяПоддержка NIH: внешняя программаНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Herricks Thurston, Donczew Magdalena, Mast Fred D, Rustad Tige, Morrison Robert, Sterling Timothy R, Sherman David R, Aitchison John D
Аннотация
PURPOSE: Timely and accurate delineation of the clinical target volume (CTV) in brachytherapy after parotid cancer surgery plays a crucial role in tailored delivery of radiation doses. This study aimed to develop and evaluate a deep learning-based model for auto-segmentation of the CTVs in postoperative adjuvant brachytherapy for patients with parotid gland cancer, addressing the challenge of achieving consistent, high-quality CTV delineations efficiently.
MATERIAL AND METHODS: Using clinical imaging data from 326 patients with parotid gland carcinoma treated at Peking University School and Hospital of Stomatology between 2017 and 2023, we established a training dataset of 213 cases, a validation set of 53 cases, and a test set of 60 cases. The CTVs on the images were segmented using 3D Res-UNet, a deep learning model, and compared against manual delineations performed by experienced radiation oncologists. The performance of 3D Res-UNet was optimized through a comprehensive preprocessing and training process tailored to the dataset's characteristics.
RESULTS: The deep learning model yielded a significant improvement in segmentation efficiency. The deep learning model generated initial CTV contours in 9.4 seconds of computational time. Subsequent expert review and minor adjustments required an average of 11.9 minutes, substantially shorter than the 46.7 minutes needed for fully manual delineation. Quantitative analysis showed that the Dice similarity coefficient (DSC) of automatic segmentation by 3D Res-UNet was 0.709, which improved to 0.924 after expert review. Qualitative evaluation by senior oncologists further affirmed the clinical acceptability of the automatically segmented CTVs.
CONCLUSIONS: Automatic contouring with physician review enabled high-accuracy and rapid CTV generation, reducing the overall delineation workload by more than 30 minutes. Consequently, the proposed deep-learning model functions as a useful support tool that streamlines postoperative adjuvant brachytherapy planning for parotid gland cancer and lessens the burden on radiation oncologists, thereby contributing to improved patient care.
MATERIAL AND METHODS: Using clinical imaging data from 326 patients with parotid gland carcinoma treated at Peking University School and Hospital of Stomatology between 2017 and 2023, we established a training dataset of 213 cases, a validation set of 53 cases, and a test set of 60 cases. The CTVs on the images were segmented using 3D Res-UNet, a deep learning model, and compared against manual delineations performed by experienced radiation oncologists. The performance of 3D Res-UNet was optimized through a comprehensive preprocessing and training process tailored to the dataset's characteristics.
RESULTS: The deep learning model yielded a significant improvement in segmentation efficiency. The deep learning model generated initial CTV contours in 9.4 seconds of computational time. Subsequent expert review and minor adjustments required an average of 11.9 minutes, substantially shorter than the 46.7 minutes needed for fully manual delineation. Quantitative analysis showed that the Dice similarity coefficient (DSC) of automatic segmentation by 3D Res-UNet was 0.709, which improved to 0.924 after expert review. Qualitative evaluation by senior oncologists further affirmed the clinical acceptability of the automatically segmented CTVs.
CONCLUSIONS: Automatic contouring with physician review enabled high-accuracy and rapid CTV generation, reducing the overall delineation workload by more than 30 minutes. Consequently, the proposed deep-learning model functions as a useful support tool that streamlines postoperative adjuvant brachytherapy planning for parotid gland cancer and lessens the burden on radiation oncologists, thereby contributing to improved patient care.