顱內出血包含自發性腦內出血及創傷性顱內出血。根據108年衛生福利部釋出的國人死因結果統計表，腦血管疾病高居第四名，其中包含缺血性中風及出血性腦中風(為自發性腦內出血的表現之一)，在臨床上必須要迅速診斷與鑑別，以進行下一步治療。在2013年，美國有280萬頭部外傷的病人至急診就診，其中必須辨識出創傷性顱內出血的病人，及早進行治療。顱內出血是非常緊急的病症，假使出血嚴重，會導致顱內的神經系統或是腦幹損傷，進而造成身體機能的受損。目前最重要的診斷工具為腦部電腦斷層，影像判讀如果可在短時間內完成，可爭取相對較多的治療時間與相應的措施。
近年來，因為深度學習的興起，現今大多的圖像識別任務可依靠深度學習來完成，且圖像的分割任務與分類任務在自然影像已有長足的進步。在分割任務中，能夠精準定位目標位置。而分類，則可以快速判斷圖像中物體的種類。延伸此概念至本研究中，腦出血可透過ＣＴ影像判讀，出血在圖像中有相應的特徵可以觀察，可視為一種圖像識別的任務，適合引入深度學習的技術，加快判讀的速度。但在一般的工作流程中，影像科醫師不只是需要判讀出血的位置，由於腦出血種類的不同，在臨床上也會有相對應的處理方法，因此必須同時區分出血腫的種類。
基于上述的工作模式與需求，我們提出了基於DenseNet的分割與分類腦出血偵測模型，透過單一模型即可得知此張CT的出血位置以及最有可能的出血類型。在現存研究中，腦出血的分割與偵測大多都是基於龐大參數量的模型或是大量資料來完成，我們透過高效率的參數運用及相對少的訓練集打造了一個輕量化且多功能的模型，提供醫師更多資訊做更精確的判讀。不僅如此，由於腦部斷層掃描是由許多張圖像所組成，醫師的判讀也是以病人為單位，我們模仿醫師的判讀模式，建構了一套演算法，可以將單張的斷層掃描的預測結果，聚合成一位病人的最終判讀結果，以利醫師做報告的撰寫或是病情的判斷。
我們希望透過此套系統，協助醫師判讀影像或是提高嚴重病患的判讀優先級，以達到縮短時間的目的。也希望在未來，在醫療資源較為不足的區域，可以透過自動化的流程，減輕臨床醫師的負擔。

Intracranial hemorrhage include spontaneous intracerebral hemorrhage and traumatic intracranial hemorrhage. According to the leading causes of death in Taiwan released by the Ministry of Health, and Welfare in 2019, cerebrovascular diseases ranked fourth. The majority of cerebrovascular diseases include ischemic stroke and hemorrhagic stroke (one of presentations of spontaneous intracerebral hemorrhage). Rapid and accurate differentiation is required for further treatment.
During 2013, there were 2.8 million visits to the emergency department because of traumatic brain injury. It is important to recognize those with traumatic intracranial hemorrhage to achieve early intervention. Intracranial hemorrhage is very urgent. Once the brain gets an injury and causes hemorrhage, blood may occupy the intracranial space to compress the organs especially the brainstem in a short and lead to the death of life. Currently, the most widely used diagnostic tool is brain computed tomography. If the interpreting of intracranial hemorrhage can be accomplished within a short time, relatively more treatment time and corresponding measures can be obtained for the patient.
In recent years, due to the rise of deep learning, most image recognition tasks can be completed by deep learning, and image segmentation tasks and classification tasks have made considerable progress in natural images. In the segmentation task, we can accurately locate the target location. And we can also know what the objects are in the image in classification.
Extending this concept we mentioned in the last paragraph to our study, intracranial hemorrhage can be interpreted through computed tomography (CT) images, and the hemorrhage can be observed in the image with corresponding characteristics. Thus, it can be regarded as an image recognition task, suitable for applying deep learning technology to accelerate the speed of interpretation. However, in a general workflow, radiologists not only find the location of hemorrhage but also need to differentiate the types of hemorrhages at the same time. According to the different types of hemorrhage, there will be corresponding treatment methods in the clinic.
According to the workflow and requirements we stated above, in this research, we propose a hemorrhage detection model based on DenseNet with multiple functional branches. Once the CT scan images are processed by the model, the location of the hemorrhage and the most likely types can be distinguished simultaneously. In the existing research, most of the approaches to the segmentation or detection tasks of intracranial hemorrhage were conducted with a single giant model or a huge training dataset. We have built a lightweight model with high-efficiency parameters, and we also used a relatively small training dataset to build a multifunctional model which can provide adequate information to radiologists for interpreting more accurately. Besides, the computed tomography scan result of the brain is composed of many images, and the radiologists interpret the possible diagnosis at the patient-level. As a result, we imitated the workflow of the radiologists and constructed a set of algorithms that can aggregate the predicted results of the CT slides to the final interpretation of a patient.
We hope that this set of algorithms can assist radiologists in interpreting images or increase the interpretation priority of serious patients to shorten the waiting time. We also hope that in the future, in the community with relatively insufficient medical resources, this automated process can be used to reduce the burden of the clinical physician.

108年度死因統計- 統計處.
Badrinarayanan, V., Kendall, A., & Cipolla, R. "SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation". IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(12), 2481–2495, 2017.
Chen, L., Bentley, P., Mori, K., Misawa, K., Fujiwara, M., & Rueckert, D. "Self-supervised learning for medical image analysis using image context restoration". Medical Image Analysis, 58, 101539, 2019.
Cho, J., Park, K. S., Karki, M., Lee, E., Ko, S., Kim, J. K., … Park, S. "Improving Sensitivity on Identification and Delineation of Intracranial Hemorrhage Lesion Using Cascaded Deep Learning Models". Journal of Digital Imaging, 32(3), 450–461, 2019.
He, K., Zhang, X., Ren, S., & Sun, J. "Deep residual learning for image recognition". Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2016-Decem, 770–778, 2016.
Hounsfield scale - Wikipedia.
Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. Q. "Densely connected convolutional networks". Proceedings - 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, 2017-Janua, 2261–2269, 2017.
Jia Deng, Wei Dong, Socher, R., Li-Jia Li, Kai Li, & Li Fei-Fei. "ImageNet: A large-scale hierarchical image database". 248–255, 2009.
Krizhevsky, A., Sutskever, I., & Hinton, G. E. "ImageNet Classification with Deep Convolutional Neural Networks Alex". Advances in Neural Information Processing Systems 24: 25th Annual Conference on Neural Information Processing Systems 2012, NIPS 2012, 1–1432, 2007.
Kuo, W., Hӓne, C., Mukherjee, P., Malik, J., & Yuh, E. L. "Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning". Proceedings of the National Academy of Sciences of the United States of America, 116(45), 22737–22745, 2019.
Lakhani, P., & Sundaram, B. "Deep learning at chest radiography: Automated Classification of Pulmonary Tuberculosis by Using Convolutional Neural Networks". Radiology, 284(2), 574–582, 2017.
Lee, H., Yune, S., Mansouri, M., Kim, M., Tajmir, S. H., Guerrier, C. E., … Do, S. "An explainable deep-learning algorithm for the detection of acute intracranial haemorrhage from small datasets". Nature Biomedical Engineering, 3(3), 173–182, 2019.
Li, X., Lai, T., Wang, S., Chen, Q., Yang, C., & Chen, R. "Feature pyramid networks for object detection". Proceedings - 2019 IEEE Intl Conf on Parallel and Distributed Processing with Applications, Big Data and Cloud Computing, Sustainable Computing and Communications, Social Computing and Networking, ISPA/BDCloud/SustainCom/SocialCom 2019, 1500–1504, 2019.
Meninges-en - Dura mater - Wikipedia.
Oktay, O., Schlemper, J., Le Folgoc, L., Lee, M., Heinrich, M., Misawa, K., … Rueckert, D. "Attention U-Net: Learning where to look for the pancreas". ArXiv, (Midl), 2018.
Pang, J., Chen, K., Shi, J., Feng, H., Ouyang, W., & Lin, D. "Libra R-CNN: Towards balanced learning for object detection". ArXiv, (3), 821–830, 2019.
Ronneberger, O., Fischer, P., & Brox, T. "U-net: Convolutional networks for biomedical image segmentation". Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 9351, 234–241, 2015.
RSNA Intracranial Hemorrhage Detection | Kaggle.
Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. "Grad-cam: Why did you say that? visual explanations from deep networks via gradient-based localization". Revista Do Hospital Das Cl??Nicas, 17, 331–336, 2016.
Shelhamer, E., Long, J., & Darrell, T. "Fully Convolutional Networks for Semantic Segmentation". IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(4), 640–651, 2017.
Tan, M., & Le, Q. V. "EfficientNet: Rethinking model scaling for convolutional neural networks". 36th International Conference on Machine Learning, ICML 2019, 2019-June, 10691–10700, 2019.
Taylor, C. A., Bell, J. M., Breiding, M. J., & Xu, L. "Traumatic brain injury-related emergency department visits, hospitalizations, and deaths - United States, 2007 and 2013". MMWR Surveillance Summaries, 66(9), 1–16, 2017.
Ye, H., Gao, F., Yin, Y., Guo, D., Zhao, P., Lu, Y., … Xia, J. "Precise diagnosis of intracranial hemorrhage and subtypes using a three-dimensional joint convolutional and recurrent neural network". European Radiology, 29(11), 6191–6201, 2019.
Yu, K. H., Zhang, C., Berry, G. J., Altman, R. B., Ré, C., Rubin, D. L., & Snyder, M. "Predicting non-small cell lung cancer prognosis by fully automated microscopic pathology image features". Nature Communications, 7, 1–10, 2016.