The Saudi dental journal · 2022
Development of a deep learning model for automatic localization of radiographic markers of proposed dental implant site locations.
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.
Alsomali Mona, Alghamdi Shatha, Alotaibi Shahad, Alfadda Sara, Altwaijry Najwa, Alturaiki Isra, Al-Ekrish Asma'a
Аннотация
OBJECTIVES: To develop a Deep Learning Artificial Intelligence (AI) model that automatically localizes the position of radiographic stent gutta percha (GP) markers in cone beam computed tomography (CBCT) images to identify proposed implant sites within the images, and to test the performance of the newly developed AI model.
MATERIALS AND METHODS: Thirty-four CBCT datasets were used for initial model training, validation and testing. The CBCT datasets were those of patients who had a CBCT examination performed wearing a radiographic stent for implant treatment planning. The datasets were exported in Digital Imaging and Communications in Medicine (DICOM), then imported into the software Horos ®. Each GP marker was manually labelled for object detection and recognition by the deep learning model by drawing rectangles around the GP markers in all axial images, then the labelled images were split into training, validation, and test sets. The axial sections of 30 CBCT datasets were randomly divided into training and validation sets. four CBCT datasets were used for testing the performance of the deep learning model. Descriptive statistics were calculated for the number of GP markers present, number of correct and incorrect identifications of GP markers.
RESULT: The AI model had an 83% true positive rate for identification of the GP markers. Of the areas labelled by the AI model as GP markers, 28 % were not truly GP markers, but the overall false positive rate was 2.8 %.
CONCLUSION: An AI model for localization of GP markers in CBCT images was able to identify most of the GP markers, but 2.8% of the results were false positive and 17% were missed GP markers. Using only axial images for training an AI program is not enough to give an accurate AI model performance.
MATERIALS AND METHODS: Thirty-four CBCT datasets were used for initial model training, validation and testing. The CBCT datasets were those of patients who had a CBCT examination performed wearing a radiographic stent for implant treatment planning. The datasets were exported in Digital Imaging and Communications in Medicine (DICOM), then imported into the software Horos ®. Each GP marker was manually labelled for object detection and recognition by the deep learning model by drawing rectangles around the GP markers in all axial images, then the labelled images were split into training, validation, and test sets. The axial sections of 30 CBCT datasets were randomly divided into training and validation sets. four CBCT datasets were used for testing the performance of the deep learning model. Descriptive statistics were calculated for the number of GP markers present, number of correct and incorrect identifications of GP markers.
RESULT: The AI model had an 83% true positive rate for identification of the GP markers. Of the areas labelled by the AI model as GP markers, 28 % were not truly GP markers, but the overall false positive rate was 2.8 %.
CONCLUSION: An AI model for localization of GP markers in CBCT images was able to identify most of the GP markers, but 2.8% of the results were false positive and 17% were missed GP markers. Using only axial images for training an AI program is not enough to give an accurate AI model performance.