臺灣博碩士論文加值系統

顳葉腦疝(uncal herniation)壓迫腦幹，會造成病患意識改變、光反射消失、生命徵象紊亂等症狀，若不及早介入預後差。緊急開刀減壓，能改善腦疝與較佳的預後。比較常見的腦疝原因為外傷造成的顱內出血。頭部電腦斷層是第一線的診斷工具，顳葉腦疝影像特徵有：顳葉海馬回、鉤回被擠入小腦幕裂孔內，基底腦池因腦疝壓迫空間減少，腦幹形狀改變等。藉由這些影像特徵，醫師能及早正確的診斷病患有無顳葉腦疝，做出適當處置。
本研為回溯性研究，收案之影像期間為 1/1/2000-1/1/2020，總計標註402例有顳葉腦疝及260例正常腦部電腦斷層影像，用以設計機器學習模型，對顳葉腦疝做出自動診斷。研究流程為四部分；(1)蒐集腦疝影像資料、(2)對腦疝影像進行標注、(3) 使用神經網路convolution neural network (CNN) You Only Look Once version 3 (YOLOv3)訓練偵測腦疝模型、(4)計算準確率。研究結果YOLOv3在顳葉腦疝的邊界偵測上的準確度為sensitivity 0.887, specificity 0.862, positive predictive value (PPV) 0.878, negative predictive value (NPV) 0.854, accuracy rate 0.824，顯示機器學習應用在顳葉腦疝的辨識是可行的。

Uncal herniation, as known as one type of descending transtentorial herniation, compresses the brain stem, results in conscious changes, loss of light reflex, contralateral central hemiplegia, and vital sign disturbances. The causes of brain herniation include space-occupying lesions such as intracranial hemorrhage, brain tumors, ischemic stroke, and cerebral edema. Accurate diagnosis and early treatment of the uncal herniation are important. Head computer tomography (CT) is the first-line diagnostic tool. The characteristics of uncal herniation on CT images can be identified by the experienced doctor for appropriate treatment. The purpose of the project is to develop a computer-aided detection and diagnosis system for uncal herniation. The retrospective study was conducted with institutional review board approval. The noncontrast-enhanced head CT examinations include 402 cases with uncal herniation and 260 cases without uncal herniation in hospitals of Kaohsiung Medical University, resulting in total 3195 representative images. The study dates of these images are between 1/1/2000-1/1/2020. The images were labeled by radiologists on CT images and were used to train and validation the model of the convolutional neural network (CNN) YOLO v3. The testing result shows good accuracy with sensitivity 0.887, specificity 0.862, PPV 0.878, NPV 0.854 and accuracy rate 0.824. The inference model with one of the deep learning algorithm, YOLO, will be a promising approach to solve delayed and missed diagnosis of uncal herniation.

論文中文摘要.............1
論文英文摘要.............2
致謝辭..................3
目次....................4
論文正文 ................5
一、前言................5
二、前人研究............7
三、研究材料與方法.......8
四、研究結果............11
五、討論...............13
六、結論...............17
參考文獻................18
附錄....................19

1.Cadena R., et al. Emergency Neurological Life Support: Intracranial Hypertension and Herniation. Neurocrit Care. 2017 Sep;27(Suppl 1):82-88.
2.Greenberg MS. Handbook of Neurosurgery 8th. New York, New York: Thieme; 2016
3.A G Osborn. Diagnosis of descending transtentorial herniation by cranial computed tomography. Radiology. 1977 Apr;123(1):93-6.
4.Gilardi, B.R., et al. Types of Cerebral Herniation and Their Imaging Features. RadioGraphics 2019; 39:1598–1610
5.Hsu, I-L, et al. An Epidemiological Analysis of Head Injuries in Taiwan. Int. J. Environ. Res. Public Health 2018, 15, 2457
6.Gurer, B., et al. The Surgical Outcome of Traumatic Extraaxial Hematomas Causing Brain Herniation. Turk Neurosurg. 2017;27(1):37-52.
7.Skoglund, TS, Nellgaˆrd, B. Long-time outcome after transient transtentorial herniation in patients with traumatic brain injury. Acta Anaesthesiol Scand. 2005;49(3):337–40.
8.Prevedello, L.M., et al. Automated critical test findings identification and online notification system using artificial intelligence in imaging. Radiology, 285(3):923–931, 2017.
9.Chilamkurthy, S., et al. Development and Validation of Deep Learning Algorithms for Detection of Critical Findings in Head CT Scans
10.Redmon, J., et al. You Only Look Once: Unified, Real-Time Object Detection