Index
Entwicklung von Prozessabläufen für die Forschungszusammenarbeit in datengetriebenen und institutionsübergreifenden Forschungsprojekten
Der vorliegende Artikel adressiert die Entwicklung von Prozessabläufen in institutionsübergreifenden und datengetriebenen Forschungsprojekten. Hierbei wird die Frage behandelt, ob eine Standardisierung der Prozessketten möglich ist, inwiefern Governance-Strukturen der Medizininformatikinitiative (MII) für datengetriebene Forschungsvorhaben inkludiert werden können und abschließend, ob somit die Handlungssicherheit für Beteiligte erhöht werden kann. Hierfür wurden Ist-Abläufe innerhalb durchgeführter Kooperationen mit empfohlenen Standards der MII abgeglichen und mithilfe des Knowhows beteiligter Mitarbeiter in Prozessketten überführt. Es konnten so Prozessabläufe entwickelt werden, die durch kaskadierende Prozessketten, Erläuterungen und Checklisten eine standardisierte Handreichung für Kooperationsprojekte bilden. Ebenfalls können durch die Dokumente zukünftig Fehler innerhalb der einzelnen Prozesselemente vermieden werden und Kooperationsprojekte einfacher, zielorientierte und übersichtlicher durchgeführt werden.
Detection of Pulmonary Embolisms in NCCT Data using Deep Learning Methods
Thesis Description
Pulmonary embolism (PE) is the third most common case of cardiovascular disease, after myocardial infection and stroke [1]. According to the report of the American Heart Association in 2020, there were 370,000 cases of PE in the US in 2016 [2]. PE most commonly originate from a deep venous thrombosis in the legs, which breaks free and migrates towards the heart and lodges in the pulmonary arteries [3]. The 30-day mortality rate for PE in the US was 9.1% in 2020 [2]. Hereby, the majority of cases of preventable deaths occur due to missed diagnosis, but not due to failure of therapy [4]. Therefore, a rapid and correct diagnosis can reduce the fatality rate and is crucial for the patient’s prognosis [5, 6].
In the context of diagnostic tools for PE, computed tomographic pulmonary angiography (CTPA) has become the gold standard imaging test, as the pulmonary filling defect can be visualized with the help of contrast agent [5]. However, CTPA does not provide functional information on lung perfusion and clots in more distal branches of the lung arteries are hard to detect [7]. A rather new method, dual energy CT (DECT), counteracts these problems, as it is possible to visualize the iodine distribution in the lung parenchyma with the help of iodine maps for functional assessment of lung perfusion [8]. The use of a dual-source CT scanner allows for simultaneous image acquisition at two different energy enabling a decomposition into iodine and non-iodine, which is not possible in plain CTPA [7, 9]. The decomposition helps to understand hemodynamics in the parenchyma of interest, here the pulmonary vascular tree. However, there are drawbacks of DECT scans for PE diagnosis. A common reason for misdiagnosis of PE is the occurrence of heterogeneous perfusion due to beam-hardening artifacts from high concentration of contrast agent [7, 9]. Another problem is the accessibility of special scanners with two X-ray sources, that can execute the DECT.
In general, the performance of a CT scan with contrast agent is a time-consuming and cumbersome procedure that might require special scanners, which may not be available in every part of the world. Additionally, patients are asked to hold their breath during scan time, which can be infeasible considering children or patients that are not fully conscious [10, 11]. A non-contrast CT scan (NCCT) can be performed within seconds and is a less difficult procedure. The detection of PE in a NCCT scan is difficult but possible. Several studies showed that especially central PE could be identified in unenhanced CT scans [12, 13]. Yet, the evaluation of NCCT scans by radiologists is neither sensitive nor specific enough to reliably detect PE [14].
To guide and speed up the physicians diagnosis of PE, machine learning algorithms have been used to automatically detect PE in current research [15, 16]. These methods were developed with the intent to identify clinically important PE and prioritize worklists, considering the increasing number of e.g. CTPA scans [15]. Weikert et al. evaluated the performance of a two-staged AI-powered algorithm detecting PE from CTPA scans [15]. The algorithm includes a 3D deep convolutional neural network based on a ResNet architecture. Evaluation of the algorithm with the institution’s test set (n=1499) resulted in a sensitivity of 92.7% and specificity of 95.5% for detecting PE. Another model, PENet, was developed by Hunag et al., a 77-layer 3D convolutional neural network (CNN), evaluated on data from 2 different institutions [16]. Detecting PE, it achieved an AUROC of 0.84 on the internal test set and 0.85 on the external dataset, using the entire volumetric CTPA imaging data.
Under the aspect of availability and feasibility of contrast-enhanced CT scans, the automatic detection of PE using a NCCT scan is desirable. Additionally, modern deep learning methods could possibly exceed human performance in the detection of PE in unenhanced CT scans. Other work has already proven the success of AIbased detection of PE in CTPA data [17], whereas it is now to be investigated if the automated detection based on unenhanced CT scans is possible. Therefore, the main goal of this work is to detect PE in an NCCT scan. This thesis explores a recently published method for medical object detection, nnDetection, which is supposed to adapt itself to arbitrary detection problems [18]. Furthermore, depending on its performance on the given task, this method is extended, or other deep learning architectures are exploited alongside nnDetection. These consider the topology of the lung tree, in order to tackle the goal of PE detection in NCCT images. In addition, this work includes a detailed analysis of the models’ performances accomplishing the task of PE detection. In general, the thesis covers the following topics:
- Systematic literature research
- Detection of PE with Deep Learning
- nnDetection as baseline
- Extension of the baseline, by further improvements using prior knowledge of data and/or
- Development of other suitable Deep Learning methods
- Analysis of the Deep Learning model
- Evaluation of the results
- Comparison with baseline model
References
[1] Meredith Turetz, Andrew T. Sideris, Oren A. Friedman, Nidhi Triphathi, and James M. Horowitz. Epidemiology, Pathophysiology, and Natural History of Pulmonary Embolism. Seminars in Interventional
Radiology, 35(2):92–98, 2018.
[2] Salim S. Virani, Alvaro Alonso, Emelia J. Benjamin, Marcio S. Bittencourt, Clifton W. Callaway, April P. Carson, Alanna M. Chamberlain, Alexander R. Chang, Susan Cheng, Francesca N. Delling, Luc Djousse,
and et al. Heart disease and stroke statistics – 2020 update: A report from the american heart association. Circulation, 141(9):e139–e596, 2020.
[3] S. Takach Lapner and C. Kearon. Diagnosis and management of pulmonary embolism. BMJ (Online), 346(7896):1–9, 2013.
[4] Peter F. Fedullo and Victor F. Tapson. The Evaluation of Suspected Pulmonary Embolism. New England Journal of Medicine, 349(13):1247–1256, 2003.
[5] Waleed Abdellatif, Mahmoud Ahmed Ebada, Souad Alkanj, Ahmed Negida, Nicolas Murray, Faisal Khosa, and Savvas Nicolaou. Diagnostic Accuracy of Dual-Energy CT in Detection of Acute Pulmonary Embolism: A Systematic Review and Meta-Analysis. Canadian Association of Radiologists Journal, 72(2):285–292, 2021.
[6] Cecilia Becattini, Maria Cristina Vedovati, and Giancarlo Agnelli. Diagnosis and prognosis of acute pulmonary embolism: Focus on serum troponins. Expert Review of Molecular Diagnostics, 8(3):339–349,
2008.
[7] Long Jiang Zhang, Chang Sheng Zhou, U. Joseph Schoepf, Hui Xue Sheng, Sheng Yong Wu, Aleksander W. Krazinski, Justin R. Silverman, Felix G. Meinel, Yan E. Zhao, Zong Jun Zhang, and Guang Ming Lu. Dualenergy CT lung ventilation/perfusion imaging for diagnosing pulmonary embolism. European Radiology, 23(10):2666–2675, 2013.
[8] Dong Jin Im, Jin Hur, Kyung Hwa Han, Hye Jeong Lee, Young Jin Kim, Woocheol Kwon, and Byoung Wook Choi. Acute pulmonary embolism: Retrospective cohort study of the predictive value of perfusion defect volume measured with dual-energy CT. American Journal of Roentgenology, 209(5):1015–1022, 2017.
[9] Guang Ming Lu, S. Y. Wu, B. M. Yeh, and L. J. Zhang. Dual-energy computed tomography in pulmonary embolism. British Journal of Radiology, 83(992):707–718, 2010.
[10] Marc Rodger and Philip S. Wells. Diagnosis of pulmonary embolism. Thrombosis Research, 103(6):225–238, 2001.
[11] Konstantin Nikolaou, Sven Thieme, Wieland Sommer, Thorsten Johnson, and Maximilian F. Reiser. Diagnosing pulmonary embolism: New computed tomography applications. Journal of Thoracic Imaging,
25(2):151–160, 2010.
[12] Christopher Thom and Nathan Lewis. Never say never: Identification of acute pulmonary embolism on non-contrast computed tomography imaging. American Journal of Emergency Medicine, 35(10):1584.e1– 1584.e3, 2017.
[13] Rocco Cobelli, Maurizio Zompatori, Giovanni De Luca, Gianfranco Chiari, Paolo Bresciani, and Carla Marcato. Clinical usefulness of computed tomography study without contrast injection in the evaluation
of acute pulmonary embolism. Journal of Computer Assisted Tomography, 29(1):6–12, 2005.
[14] Simon Sun, Alexandre Semionov, Xuanqian Xie, John Kosiuk, and Benoˆıt Mesurolle. Detection of central pulmonary embolism on non-contrast computed tomography: A case control study. International Journal of Cardiovascular Imaging, 30(3):639–646, 2014.
[15] Thomas Weikert, David J. Winkel, Jens Bremerich, Bram Stieltjes, Victor Parmar, Alexander W. Sauter, and Gregor Sommer. Automated detection of pulmonary embolism in CT pulmonary angiograms using an AI-powered algorithm. European Radiology, 30(12):6545–6553, 2020.
[16] Shih Cheng Huang, Tanay Kothari, Imon Banerjee, Chris Chute, Robyn L. Ball, Norah Borus, Andrew Huang, Bhavik N. Patel, Pranav Rajpurkar, Jeremy Irvin, Jared Dunnmon, Joseph Bledsoe, Katie Shpanskaya, Abhay Dhaliwal, Roham Zamanian, Andrew Y. Ng, and Matthew P. Lungren. PENet—a scalable deep-learning model for automated diagnosis of pulmonary embolism using volumetric CT imaging. npj Digital Medicine, 3(1):1–9, 2020.
[17] Shelly Soffer, Eyal Klang, Orit Shimon, Yiftach Barash, Noa Cahan, Hayit Greenspana, and Eli Konen. Deep learning for pulmonary embolism detection on computed tomography pulmonary angiogram: a systematic review and meta-analysis. Scientific Reports, 11(1):1–8, 2021.
[18] Michael Baumgartner, Paul F. J¨ager, Fabian Isensee, and Klaus H. Maier-Hein. nnDetection: A Selfconfiguring Method for Medical Object Detection. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 12905 LNCS:530–539, 2021.
Enhancing the robustness and efficiency of multimodal emotion estimation models
Representation Learning with Partial Medical Volumes
Integration of Augmented Reality in SPECT-CT Workflows
Graph Augmentation using Cond.-GANs
Post-Processing of DTF-Skeletonizations
Detection and Prediction of Background Parenchymal Enhancement on MRI Using Neural Network”
Learnable Feature Space Reductions for Acoustic Representation Vectors
Detection of Large Vessel Occlusions using Graph Deep Learning
Thesis Description
Acute ischemic stroke (AIS) is among the leading causes of death (9%) and disability worldwide and will increase over the next 20 years [1, 2]. One of the main causes for an AIS is large vessel occlusion (LVO),accounting for more than one-third of all AIS cases [3]. Thus, making it a key component in the diagnostic process and treatment [4]. These occlusions usually occurs in the internal carotid artery, proximal middle cerebral artery or basilar artery. As a treatment for AIS thrombolysis has been used over the years in a cautious manner, due to the limited application time window and the need to asses the risks to benefits ratio individually. An alternative preferred treatment method is the mechanical thrombectomy, where patients are treated surgically [5, 6]. Thereby, doing a fast classification and possibly a detection of LVOs is crucial for a successful diagnosis and thus for the treatment of AIS [4, 6].
Computed Tomography Angiography (CTA) represents one of the most important techniques in order to detect LVOs leading to the diagnosis and treatment management of AIS, by visualizing the cerebrovascular vessel tree [7]. Detecting such LVOs manually is very time consuming, especially when handling big datasets. Hence an automated classification of patients suffering from LVOs is beneficial for the clinical workflow, by shortening the diagnostic time and improving on accurate classification. For this purpose multiple algorithms were implemented and tested for such application using simple machine learning methods, such as regularized logistic regression, linear support vector machines, and random forest, providing good results as presented by Nishi et al. in [8]. In the last couple of years more sophisticated methods like deep learning, in particular convolutional neural networks (CNNs) as seen in [9, 10] are applied and improve previous algorithms. Still traditional deep learning tends to fail to modulate irregular complex structures, such as cerebrovascular vessel trees. This is mainly due to the fact, that the topology differs between patients [11]. Thus, a plausible alternative should be tested.
Graph deep learning has emerged in recent years as a powerful tool in machine learning, using graphs to model geometrical and topological relationships. It provides an advantage over traditional deep learning regarding irregular structures, such as cerebrovascular vessel trees [11]. Another reason for cosidering GDL for detecting LVOs besides the deformations invariance is the dimension reduction of the inputdata and the ability not to use the Volume as a whole. GDL also identifies and explores a wide range of parallelization strategies using simpler neural networks compared to traditional deep learning, leading to speed and memory efficiency [12, 13].
This thesis aims to create a GDL model to automatically detect LVOs, by analyzing the topological structure using distance features to classify them into LVO left, right or no LVO. This approach will be directly
compared to a state-of-the-art LVO CNN model, like in Thamm et al [9]. In summary, the thesis deals with the following points:
- Preparation of distance features for vessel graph
- Training of GDL models to analyze the graph topology
- Implementation of vessel-tree specific augmentation techniques
- Analysis and evaluation of the GDL models
References
[1] Walter Johnson, Oyere Onuma, Mayowa Owolabi, and Sonal Sachdev. Stroke: a global response is needed.
Bulletin of the World Health Organization, 94(9):634–634A, 2016.
[2] Geoffrey A. Donnan, Marc Fisher, Malcolm Macleod, and Stephen M. Davis. Stroke. The Lancet,
371(9624):1612–1623, 2008.
[3] Konark Malhotra, Jeffrey Gornbein, and Jeffrey L. Saver. Ischemic Strokes Due to Large-Vessel Occlusions
Contribute Disproportionately to Stroke-Related Dependence and Death: A Review. Frontiers in neurology,
8:1–5, 2017.
[4] Nikita Lakomkin, Mandip Dhamoon, Kirsten Carroll, Inder Paul Singh, Stanley Tuhrim, Joyce Lee, Johanna T. Fifi, and J. Mocco. Prevalence of large vessel occlusion in patients presenting with acute ischemic
stroke: a 10-year systematic review of the literature. Journal of neurointerventional surgery, 11(3):241–245,
2019.
[5] Murugan Palaniswami and Bernard Yan. Mechanical Thrombectomy Is Now the Gold Standard for Acute
Ischemic Stroke: Implications for Routine Clinical Practice. Interventional neurology, 4(1-2):18–29, 2015.
[6] Salwa El Tawil and Ketih W Muir. Thrombolysis and thrombectomy for acute ischaemic stroke. Royal
College of Physicians, pages 161–165, 2017.
[7] M. D. Ralf W. Baumgartner and et al. Transcranial color–coded duplex sonography, magnetic resonance
angiography, and computed tomography angiography: Methods, applications, advantages, and limitations.
Journal of Clinical Ultrasound, pages 89–111, 2005.
[8] M. D. Hidehisa Nishi, Naoya Oishi, MD, PhD, Akira Ishii, MD, PhD, and et al. Predicting Clinical
Outcomes of Large Vessel Occlusion Before Mechanical Thrombectomy Using Machine Learning. American
Heart Association Journals, pages 2379–2388, 2019.
[9] Florian Thamm, Oliver Taubmann, Markus J¨urgens, Hendrik Ditt, and Andreas Maier. Detection of Large
Vessel Occlusions using Deep Learning by Deforming Vessel Tree Segmentations. EESS.IV, pages 1–7,
2021.
[10] Lucas W. Remedios, Sneha Lingam, Samuel W. Remedios, Riqiang Gao, Stephen W. Clark, Larry T.
Davis, and Bennett A. Landman. Comparison of convolutional neural networks for detecting large vessel
occlusion on computed tomography angiography. Medical physics, 48(10):6060–6068, 2021.
[11] Stavros Georgousis, Michael P. Kenning, and Xianghua Xie. Graph Deep Learning: State of the Art and
Challenges. IEEE Access, 9:22106–22140, 2021.
[12] Si Zhang, Hanghang Tong, Jiejun Xu, and Ross Maciejewski. Graph convolutional networks: a comprehensive review. Computational Social Networks, 6(1):626, 2019.
[13] Federico Monti, Davide Boscaini, and et al. Geometric Deep Learning on Graphs and Manifolds Using
Mixture Model CNNs. IEEE Xplore, (2017):5115–5124.