Journal of applied clinical medical physics · 2022
Deep learning for emergency ascites diagnosis using ultrasonography images.
СтатьяНазвание и аннотация: Перевод в очереди
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.
Lin Zhanye, Li Zhengyi, Cao Peng, Lin Yingying, Liang Fengting, He Jiajun, Huang Libing
Аннотация
PURPOSE: The detection of abdominal free fluid or hemoperitoneum can provide critical information for clinical diagnosis and treatment, particularly in emergencies. This study investigates the use of deep learning (DL) for identifying peritoneal free fluid in ultrasonography (US) images of the abdominal cavity, which can help inexperienced physicians or non-professional people in diagnosis. It focuses specifically on first-response scenarios involving focused assessment with sonography for trauma (FAST) technique.
METHODS: A total of 2985 US images were collected from ascites patients treated from 1 January 2016 to 31 December 2017 at the Shenzhen Second People's Hospital. The data were categorized as Ascites-1, Ascites-2, or Ascites-3, based on the surrounding anatomy. A uniform standard for regions of interest (ROIs) and the lack of obstruction from acoustic shadow was used to classify positive samples. These images were then divided into training (90%) and test (10%) datasets to evaluate the performance of a U-net model, utilizing an encoder-decoder architecture and contracting and expansive paths, developed as part of the study.
RESULTS: Test results produced sensitivity and specificity values of 94.38% and 68.13%, respectively, in the diagnosis of Ascites-1 US images, with an average Dice coefficient of 0.65 (standard deviation [SD] = 0.21). Similarly, the sensitivity and specificity for Ascites-2 were 97.12% and 86.33%, respectively, with an average Dice coefficient of 0.79 (SD = 0.14). The accuracy and area under the curve (AUC) were 81.25% and 0.76 for Ascites-1 and 91.73% and 0.91 for Ascites-2.
CONCLUSION: The results produced by the U-net demonstrate the viability of DL for automated ascites diagnosis. This suggests the proposed technique could be highly valuable for improving FAST-based preliminary diagnoses, particularly in emergency scenarios.
METHODS: A total of 2985 US images were collected from ascites patients treated from 1 January 2016 to 31 December 2017 at the Shenzhen Second People's Hospital. The data were categorized as Ascites-1, Ascites-2, or Ascites-3, based on the surrounding anatomy. A uniform standard for regions of interest (ROIs) and the lack of obstruction from acoustic shadow was used to classify positive samples. These images were then divided into training (90%) and test (10%) datasets to evaluate the performance of a U-net model, utilizing an encoder-decoder architecture and contracting and expansive paths, developed as part of the study.
RESULTS: Test results produced sensitivity and specificity values of 94.38% and 68.13%, respectively, in the diagnosis of Ascites-1 US images, with an average Dice coefficient of 0.65 (standard deviation [SD] = 0.21). Similarly, the sensitivity and specificity for Ascites-2 were 97.12% and 86.33%, respectively, with an average Dice coefficient of 0.79 (SD = 0.14). The accuracy and area under the curve (AUC) were 81.25% and 0.76 for Ascites-1 and 91.73% and 0.91 for Ascites-2.
CONCLUSION: The results produced by the U-net demonstrate the viability of DL for automated ascites diagnosis. This suggests the proposed technique could be highly valuable for improving FAST-based preliminary diagnoses, particularly in emergency scenarios.