Intelligent Systems in Medical Imaging
Lecturer: Francesco Ciompi
Contents:
W1: Basic Techniques
- ISMI-1a-Introduction_Medical_Imaging.md
- ISMI-1b-Digital_Image_Formation.md
- ISMI-1c-Intensity_Transformation.md
- ISMI-1d-Spatial_Filtering.md
- "Types of filters I included with this one"
W2: Segmentation (old techniques)
Suggested paper: Comparative study of retinal vessel segmentation methods on a new publicly available database
- ISMI-2a-Sementation_Detection_Classification.md
- ISMI-2b-Segmentation.md
- ISMI-2c-Binary_morphology.md
- ISMI-2d-Classification.md
- ISMI-2e-Detection
Paper:
W3: Introduction supervised learning
TODO Suggested reading: Lecun et al. - Gradient-Based Learning Applied to Document Recognition (only first pages)
W4: Convolutional Networks Basic Techniques
Suggested paper: ImageNet Classification with Deep Convolutional Neural Networks
- ISMI-4a-Short_history_deep_learning_revolution
- ISMI-4b-Convolutional_Neural_Networks
- ISMI-4c-CNN_network_parameters
- ISMI-4d-ReLu_with_He_initialization
- ISMI-4e-Regularization
- ISMI-4f-Batch_normalization
- ISMI-4g-Update_rules
- ISMI-4h-Convolutional_nets_in_practice
W5: Convolutional networks (for segmentation)
Suggested papers:
- Multi-Scale Context Aggregation by Dilated Convolutions, Fisher Yu & Vladlen Koltun.
-
U-Net: Convolutional Network for Biomedical Image Segmentation; Ronneberger, Fisher, Brox.
- ISMI-5b-Fully_convolutional_nets
- ISMI-5c-Dilated_convolutions
- ISMI-5d-Upscaling_feature_maps
- ISMI-5e-Unet
- ISMI-5f-Popular_CNN_architectures
W6: Object Detection using CNNs
Suggested papers:
- Rich feature hierarchies for accurate object detection and semantic segmentation - by Ross Girschick et al.
- First paper about R-CNN
-
You Only Look Once: Unified, Real-Time Object Detection, by Joseph Redmon et al.
- ISMI-6b-Object_Detection_Sampling_strategies
- Jaccard_Index_intersection_over_union.md
- ISMI-6d-Region_proposal_networks
- I find Faster R-CNN very vaguely explained
- ISMI-6e-Yolo
- T.B.C. vond het weer vaag
Exam preparation
TODO: read some papers?
- Be able to recognize different types of medical imaging problems
- For those problems, be able to select appropriate techniques
- Be able to name the amount of parameters for common layers, such as a convolution layer or dense layer.
- Be able to quickly determine the output dimensions after common operations, such as convolution with a type of padding and stride
- Understand which CNN parameters are data-driven and which are design choices