Explainable deep learning models for healthcare - CDSS 3

Explainable deep learning models for healthcare - CDSS 3

This course is part of Informed Clinical Decision Making using Deep Learning Specialization

Instructor: Fani Deligianni

Access provided by Taipei Medical University & IOOOI Ally

1,851 already enrolled

4 modules

Gain insight into a topic and learn the fundamentals.

15 reviews

Intermediate level

Recommended experience

3 weeks to complete

at 10 hours a week

Flexible schedule

Learn at your own pace

4 modules

Gain insight into a topic and learn the fundamentals.

15 reviews

Intermediate level

Recommended experience

3 weeks to complete

at 10 hours a week

Flexible schedule

Learn at your own pace

What you'll learn

Program global explainability methods in time-series classification
Program local explainability methods for deep learning such as CAM and GRAD-CAM
Understand axiomatic attributions for deep learning networks
Incorporate attention in Recurrent Neural Networks and visualise the attention weights

Skills you'll gain

Tools you'll learn

Details to know

Shareable certificate

Add to your LinkedIn profile

Assessments

5 assignments

Taught in English

See how employees at top companies are mastering in-demand skills

Learn more about Coursera for Business

logos of Petrobras, TATA, Danone, Capgemini, P&G and L'Oreal

Build your subject-matter expertise

This course is part of the Informed Clinical Decision Making using Deep Learning Specialization

When you enroll in this course, you'll also be enrolled in this Specialization.

Learn new concepts from industry experts
Gain a foundational understanding of a subject or tool
Develop job-relevant skills with hands-on projects
Earn a shareable career certificate

There are 4 modules in this course

This course will introduce the concepts of interpretability and explainability in machine learning applications. The learner will understand the difference between global, local, model-agnostic and model-specific explanations. State-of-the-art explainability methods such as Permutation Feature Importance (PFI), Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanation (SHAP) are explained and applied in time-series classification. Subsequently, model-specific explanations such as Class-Activation Mapping (CAM) and Gradient-Weighted CAM are explained and implemented. The learners will understand axiomatic attributions and why they are important. Finally, attention mechanisms are going to be incorporated after Recurrent Layers and the attention weights will be visualised to produce local explanations of the model.

Deep learning models are complex and it is difficult to understand their decisions. Explainability methods aim to shed light to the deep learning decisions and enhance trust, avoid mistakes and ensure ethical use of AI. Explanations can be categorised as global, local, model-agnostic and model-specific. Permutation feature importance is a global, model agnostic explainabillity method that provide information with relation to which input variables are more related to the output.

What's included

6 videos8 readings1 assignment1 discussion prompt5 ungraded labs

6 videos Total 72 minutes

Welcome video - Explainable Deep Learning Models for Healthcare 1 minute
Interpretability vs Explainability 17 minutes
'Explainability' in Healthcare Applications 17 minutes
Taxonomy of Explainability Methods 17 minutes
Model Agnostic Explainability Methods 11 minutes
Permutation Feature Importance in Time Series Data 8 minutes

8 readings Total 390 minutes

The importance of explainable prediction models in healthcare 10 minutes
Explainable Artificial Intelligence - Taxonomy 60 minutes
Model Agnostic Explainability 30 minutes
Permutation Feature Importance 60 minutes
Practical Exercise: Interpretability of the MLP model using Permutation Feature Importanceg 60 minutes
Practical Exercise: Interpretability of the CNN model using Permutation Feature Importance 60 minutes
Practical Exercise: Interpretability of the LSTM model using Permutation Feature Importance 60 minutes
Explainability models in ECG 50 minutes

1 assignment Total 30 minutes

End of week 1 quiz 30 minutes

1 discussion prompt Total 10 minutes

Week 1 - Your experience 10 minutes

5 ungraded labs Total 50 minutes

Permutation feature importance for classifying heart beats using a CNN 10 minutes
Light - Permutation feature importance for classifying heart beats using a CNN 10 minutes
Permutation feature importance for classifying heart beats using an LSTM 10 minutes
Light - Permutation feature importance for classifying heart beats using an LSTM 10 minutes
Permutation feature importance for classifying heart beats using a multi-layer perceptron 10 minutes

Local explainability methods provide explanations on how the model reach a specific decision. LIME approximates the model locally with a simpler, interpretable model. SHAP expands on this and it is also designed to address multi-collinearity of the input features. Both LIME and SHAP are local, model-agnostic explanations. On the other hand, CAM is a class-discriminative visualisation techniques, specifically designed to provide local explanations in deep neural networks.

What's included

5 videos7 readings1 assignment1 discussion prompt7 ungraded labs

5 videos Total 48 minutes

Local Interpretable Model Agnostic Explanations (LIME) 9 minutes
LIME in Time-Series Classification 10 minutes
Shapley Additive Explanations 12 minutes
Model-Specific Explanations: Visualisation Methods 13 minutes
CAM in Time-Series Classification 5 minutes

7 readings Total 320 minutes

Why Should I Trust You? 30 minutes
Practical Exercise: Interpretability of heartbeat classification using LIME and an NNMLP model 60 minutes
Practical Exercise: Interpretability of heartbeat classification using LIME and a CNN model 60 minutes
Practical Exercise: Interpretability of heartbeat classification using LIME and an LSTM model 60 minutes
A Unified Approach to Interpreting Model Predictions 40 minutes
Practical Exercise: Interpretability of CNN models using Class Activation Maps 10 minutes
Class Activation Mapping 60 minutes

1 assignment Total 30 minutes

End of week 2 quiz 30 minutes

1 discussion prompt Total 10 minutes

Week 2 - Your experience 10 minutes

7 ungraded labs Total 65 minutes

Interpretability of heartbeat classification using a CNN model and Class Activation Maps 10 minutes
Interpretability of heartbeat classification using a CNN model and CAM 10 minutes
LIME interpretability for heartbeat classification with a convolutional neural network 10 minutes
Interpretability of heartbeat classification using an LSTM model and CAM 10 minutes
LIME interpretability for heartbeat classification with a long short-term memory network 10 minutes
Light - LIME interpretability for heartbeat classification with a long short-term memory network 5 minutes
LIME interpretability for heartbeat classification with a multi-layer perceptron 10 minutes

GRAD-CAM is an extension of CAM, which aims to a broader application of the architecture in deep neural networks. Although, it is one of the most popular methods in explaining deep neural network decisions, it violates key axiomatic properties, such as sensitivity and completeness. Integrated gradients is an axiomatic attribution method that aims to cover this gap.

What's included

4 videos6 readings1 assignment1 discussion prompt7 ungraded labs

4 videos Total 37 minutes

Gradient Weighted Class Activation Maps 12 minutes
Grad-CAM in Time-Series Classification 9 minutes
Integrated Gradients 12 minutes
Integrated Gradients in Time Series Classification 4 minutes

6 readings Total 310 minutes

GRAD - Class Activation Mapping 30 minutes
Practical Exercise: Interpretability of the CNN model using Gradient-weighted Class Activation Mapping 60 minutes
Practical Exercise: Interpretability of the LSTM model using Gradient-weighted Class Activation Mapping 60 minutes
Axiomatic Attribution for Deep Networks 40 minutes
Practical Exercise: Interpretability of the CNN model using Integrated Gradients 60 minutes
Practical Exercise: Interpretability of the LSTM model using Integrated Gradients 60 minutes

1 assignment Total 30 minutes

End of week 3 quiz 30 minutes

1 discussion prompt Total 10 minutes

Week 3 - Your experience 10 minutes

7 ungraded labs Total 70 minutes

Interpretability of heartbeat classification using a CNN model and Grad-CAM 10 minutes
Interpretability of heartbeat classification using integrated gradients and a CNN model 10 minutes
Light - Interpretability of heartbeat classification using integrated gradients and a CNN model 10 minutes
Interpretability of heartbeat classification using an LSTM model and Grad-CAM 10 minutes
Light - Interpretability of heartbeat classification using an LSTM model and Grad-CAM 10 minutes
Interpretability of heartbeat classification using integrated gradients and an LSTM model 10 minutes
Light - Interpretability of heartbeat classification using integrated gradients and an LSTM model 10 minutes

Attention in deep neural networks mimics human attention that allocates computational resources to a small range of sensory input in order to process specific information with limited processing power. In this week, we discuss how to incorporate attention in Recurrent Neural Networks and autoencoders. Furthermore, we visualise attention weights in order to provide a form of inherent explanation for the decision making process.

What's included

3 videos3 readings2 assignments1 discussion prompt4 ungraded labs

3 videos Total 34 minutes

Attention in Deep Learning 14 minutes
Taxonomy of Attention 13 minutes
Attention and Explainability 8 minutes

3 readings Total 190 minutes

Survey on Attention Mechanisms 10 minutes
Practical Exercise: Classification of heartbeats using an LSTM with attention mechanism 90 minutes
Practical Exercise: Interpretability of the LSTM model with attention mechanism 90 minutes

2 assignments Total 60 minutes

End of course summative quiz 30 minutes
End of week 4 quiz 30 minutes

1 discussion prompt Total 10 minutes

Week 4 - Your experience 10 minutes

4 ungraded labs Total 30 minutes

Interpretability of heartbeat classification using an LSTM model with attention mechanism 10 minutes
Light - Interpretability of heartbeat classification using an LSTM model with attention mechanism 5 minutes
Heartbeat classification using an LSTM model with attention mechanism 10 minutes
Light - Heartbeat classification using an LSTM model with attention mechanism 5 minutes