Module CS4374-KP06

Duration

1 Semester

Turnus of offer

each summer semester

Credit points

6

Course of studies, specific fields and terms:

• Master MES 2020, optional subject, computer science / electrical engineering
• Master Robotics and Autonomous Systems 2019, optional subject, Elective
• Master Medical Informatics 2014, optional subject, medical computer science
• Master MES 2014, optional subject, computer science / electrical engineering
• Master Medical Informatics 2019, advanced module, medical computer science

Classes and lectures:

• Medical Deep Learning (exercise, 2 SWS)
• Medical Deep Learning (lecture, 2 SWS)

Workload:

• 60 hours in-classroom work
• 80 hours private studies
• 40 hours exam preparation

Contents of teaching:

• Cardiac Healthcare:
• ECG signal analysis for arrhythmia detection or sleep apnea and for mobile low-cost devices
• MRI sequence analysis for anatomical segmentation and temporal modelling
• Multimodal Clinical Case Retrieval / Prediction:
• Pathology and Semantic Image Retrieval and Localisation
• Analysis of text / natural language (radiology reports/study articles) for multimodal data mining in Electronic Health Records (EHR)
• Computer Aided Detection and Disease Classification:
• CT Lung nodule detection for cancer screening with data augmentation and transfer learning
• Weakly-supervised abnormality detection and biomarker discovery
• Interpretable and reliable deep learning systems
• Human interaction and correction within deep learning models
• Visualisation of uncertainty and internally learned representations
• Deep Learning Concepts, Architectures and Hardware
• Convolutional Neural Networks, Layers, Deep Residual Learning
• Losses, Derivatives, Large-scale Stochastic Optimisation
• Directed Acyclic Graph Networks, Generative Adversarial Networks
• Cloud Computing, GPUs, Low Precision Computing, DL Frameworks

Qualification-goals/Competencies:

• Students know the importance of data security, patient anonymisation and ethics for clinical studies involving sensitive data
• They know methods and tools to collect, preprocess, store and annotate large datasets for deep learning from medical data
• They have an in-depth understanding of deep / convolutional neural networks for general data (signals / text / images) processing, their learning process and evaluation of their performance on unseen data
• They understand the principles of weakly-supervised learning, transfer learning, concept discovery and generative adversarial networks
• They know how to explore learned feature representations for retrieval and visualisation of high-dimensional abstract data
• They can implement modern network architectures in DL frameworks and are able to adapt and extend them to given problems in medicine
• They have a broad overview of current applications of deep learning in medicine in both research and clinical practice and can transfer their knowledge to newly emerging domains

Grading through:

• Oral examination

Responsible for this module:

• Prof. Dr. Mattias Heinrich

Teacher:

• Institute of Medical Informatics
• Prof. Dr. Mattias Heinrich

Literature:

• Ian Goodfellow, Yoshua Bengio and Aaron Courville : Deep Learning The MIT Press

Language:

• English, except in case of only German-speaking participants

Notes:

Last Updated:

24.09.2021