Deep learning for

image data

Karl Kumbier

March 6, 2019

Deep learning: "end to end" representations 

Convolutional neural network (CNN)

Fukushima (1980), LeCun et al. (1998), Krizhevsky (2012)

Deep learning: "end to end" representations 

Convolutional neural network (CNN)

Classification

Payan and Montana (2015)

Fukushima (1980), LeCun et al. (1998), Krizhevsky (2012)

Deep learning: "end to end" representations 

Convolutional neural network (CNN)

Classification

Payan and Montana (2015)

Segmentation

Yang et al. (2016)

Fukushima (1980), LeCun et al. (1998), Krizhevsky (2012)

Deep learning: "end to end" representations 

Convolutional neural network (CNN)

Classification

Payan and Montana (2015)

Segmentation

Yang et al. (2016)

Registration

Li and Fan (2017)

Fukushima (1980), LeCun et al. (1998), Krizhevsky (2012)

Challenges for biomedical images

• Distinct imaging modalities (multi-channel, 3D, etc.)
• Limited availability of labeled data

Transfer learning

Antony et al. (2016), Kim et al. (2016)

• Distinct imaging modalities (multi-channel, 3D, etc.)
• Limited availability of labeled data

Challenges for biomedical images

Transfer learning

Antony et al. (2016), Kim et al. (2016)

Problem formulation

Kraus et al. (2016)

• Distinct imaging modalities (multi-channel, 3D, etc.)
• Limited availability of labeled data

Challenges for biomedical images

Transfer learning

Data augmentation 

Real

Antony et al. (2016), Kim et al. (2016)

Ronneberger et al. (2015), Uzunova  et al. (2017)

Deformed

Generated

Problem formulation

Kraus et al. (2016)

• Distinct imaging modalities (multi-channel, 3D, etc.)
• Limited availability of labeled data

Challenges for biomedical images

Deep learning in the lab

• Extracting high dimensional information from microscopy data

• Hypothesis generation/discovery from high dimensional data

• Improved automation