AUTHOR=Shan Wei , Guo Jianwei , Mao Xuewei , Zhang Yulei , Huang Yikun , Wang Shuai , Li Zixiao , Meng Xia , Zhang Pingye , Wu Zhenzhou , Wang Qun , Liu Yaou , He Kunlun , Wang Yongjun 

TITLE=Automated Identification of Skull Fractures With Deep Learning: A Comparison Between Object Detection and Segmentation Approach

JOURNAL=Frontiers in Neurology

VOLUME=Volume 12 - 2021

YEAR=2021

URL=https://www.frontiersin.org/journals/neurology/articles/10.3389/fneur.2021.687931

DOI=10.3389/fneur.2021.687931

ISSN=1664-2295

ABSTRACT=Objective:
To develop a well-established system for automatic detection of skull fractures to make clinical diagnoses that would be useful and efficient in clinical practice.
Method:
This is a retrospective study based on CT data sets (CT bone algorithm sequences) acquired from 4,782 patients (median age, 54 years; 2583 males, 2199 females) between September 2016 and September 2020. All the patients had been diagnosed with skull fractures, with clear CT characteristics of skull fracture. These data were divided into training, validation, and testing cohorts comprised of 3,326 patients, 842 patients, and 614 patients, respectively. This study also included approximately 7,856 healthy people whose skull structure was normal in the training set. All these data for the model training were labeled layer by layer and by consensus by seven doctors with 6 years of work experience. Two different convolutional neural networks were applied to 4,168 training and validation CT data sets to construct two different Deep Learning Systems (DLSs), which were tested in 614 independent CT data sets for algorithm optimization. Next, these two DLS tools were used to run a skull fracture detection program in the novel clinical recruitment test data set for clinical evaluation.
Results:
In the skull fracture detection program in the novel clinical recruitment test dataset for clinical evaluation, these two methods achieved 83.83% and 92.47% skull fracture lesion detection.
Conclusion:
The results demonstrate that automatic detection of skull fractures is feasible. The well-trained DLS system could be a trusted tool for the detection of skull fractures.