Artificial Intelligence Flashcards
Define artificial intelligence
Technology that leverages computers and machines to mimic the problem-solving and decision-making capabilities of the human mind.
Define machine learning
A branch of artificial intelligence (AI) and computer science which focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy.
State 4 reasons why AI is now commercially available
- Computer power is great enough that AI models have become commercially viable
- Several models have been developed that show promise
- Commercial products have become available
- The public are gradually accepting AI as a positive development
State 9 clinical applications of AI
1) Automatic image segmentation
2) Automatic optimisation of radiotherapy planning parameters
3) Identification of images with abnormal pathology
4) Highlighting areas of interest in images
5) Prioritising images for further (human) review
6) MRI ‘acceleration’
7) Patient scheduling and bed management
8) Early warning about changes in equipment performance
9) Generating synthetic patient data
What are the 4 general classes of AI models?
- Image or data classification
- Regression (predicting the output of a given input)
- Object detection
- Generative AI
What is the Turing test?
A measurement of the maturity of an AI system.
How does an AI system pass the Turing test?
If the computer’s output is indistinguishable from that of a human, then the test passes.
What are ‘reverse Turing tests’ used for?
To distinguish between AI and humans.
What happens without a feedback loop in AI models?
For a given input, the same output will always be given.
How is an AI model trained?
The model is fed training models, which stimulate the model and produce initial outputs. Multiple learning cycles are then run to improve the model. The hyperparameters for each learning cycle (the parameters that control the learning process of the model) are saved to understand the impacts of different configurations.
What type of model were the AI tools for predicting the diagnosis and prognosis of covid-19?
Regression
What are the dangers of implementing unreliable, biased AI models in healthcare?
Unreliable predictions could cause more harm than benefit in guiding clinical decisions.
The dataset for training an AI model should always be _______ to the model for validating an AI model.
Training
Validating
Which 2 datasets should be used for AI model development?
- Training set
- Tuning set
Which 2 datasets should be used for AI model evaluation?
- Internal validation test set
- External validation test set
Describe the datasets used for AI model training (development and evaluation)
What is a confusion matrix?
A table that evaluates the performance of a classification model by comparing its predictions against the true values.
What is the recall (sensitivity) of a classification model?
The proportion of all actual positive cases that were correctly predicted as positive by the model.
What is the specificity of a classification model?
A model’s ability to correctly identify true negative cases.
What is the F1 score of a classification model?
A metric used to evaluate a models precision for true positives AND recall. It is the harmonic mean of precision and recall so penalises models that have either high precision or recall but not both.
Give the equation for the precision of a classification model for a positive predictive value
TP = true positive
FP = false positive
Give the equation for the precision of a classification model for a negative predictive value
TN = true negative
FN = false negative
Give the equation for the recall (sensitivity) of a classification model
TP = true positive
FN = false negative
Give the equation for the specificity of a classification model
TN = true negative
FP = false positive
