Wk 4 - AI in RT

Question 1

AI

Answer 1

• human intelligence level is the end goal
• visual perception
• speech recognition
• high level decision making with minimal input
• translation between languages

Question 2

ML

Answer 2

• statistical methods of learning from data
• supervised learning
• unsupervised learning
• semi-supervised learning
• reinforcement learning
• transfer learning

Question 3

DL

Answer 3

• artificial neural network
• inspired by the brain networks of neurons
• multiple layers in the network
• various architectures
• can handle large volumes of data
• includes convolutional neural networks (CNN) and generative adversarial networks (GAN)

Question 4

the hype cycle

Answer 4

clinical understanding and expectations becoming more realistic as this technology is implemented

Question 5

augmented intelligence can help us

Answer 5

• deal with large amounts of data
• find patterns and relationships previously too difficult for humans
• develop decision aids for complex situations
• remove menial repetitive tasks

Question 6

ML approaches

Answer 6

• supervised learning
• unsupervised learning
• reinforcement learning
• transfer learning

Question 7

supervised learning

Answer 7

• common approach
• uses labelled data to help the model learn relationships present
• labelled data has both the input variables and output variable of interest in the data set
• described as supervised because it required human input to label the data

Question 8

unsupervised learning

Answer 8

• uses an unlabelled data
• models are focused on finding patterns or group present
• useful for dimensionality reduction, or reducing the number of variables
• called unsupervised because it does not require human labelling outcome

Question 9

semi-supervised learning

Answer 9

• data contains some labels (typically a minority of the data)
• allows the model to learn the relationship between the input and output variables but is not limited to looking at this connection
• reduces the burden of labelling a large data set

Question 10

reinforcement learning

Answer 10

• uses a ‘trial and error’ type approach through an agent that interacts with the environment
• the agent is set a task and is guided by reward and punishment as it makes decisions on how it approaches the task
• while similar to a human approach to perform a task, the model may address a problem in a way that humans wouldn’t normally consider but could be beneficial

Question 11

transfer learning

Answer 11

• uses a model (usually neural network) trained for one task and applies it to a similar but different task
• reduces the need to re-train large and complex deep learning models
• relies on the variables in the first task to be general and relatable to the second task
• pre trained models can be used in whole or part and adapted to a new setting
• provides potential efficiencies in model development
• can be both supervised and unsupervised

Question 12

two outcomes of supervised learning

Answer 12

• classification
• regression

Question 13

classification supervised learning

Answer 13

• predicts a class outcome
  eg. yes or no, dosimetry goal (met or not met), toxicity grade (1, 2, 3, 4)

Question 14

regression supervised learning

Answer 14

• predicts a numerical outcome
  eg. OAR dose volume, QOL score, organ motion distance

Question 15

clinical applications of ML - RT simulation

Answer 15

• brain cases were used to create pseudo CT scans by mapping from diagnostic T1 and T1 + gadolinium MRI scans
• they used a 3D convolutional neural network to do this
• they compared tow different types of 3D CNNs with different architecture, and assessed the outcome using mean absolute error, gamma indices and DVHs
• they found that they were successfully able to map from the MRIs to a pCT with similar results between model types and MRI types
• DVH calculations showed that these pCTs were suitable for clinical use

Question 16

clinical applications of ML - contouring

Answer 16

• a comparison of manual and deep learning based contours
• CTVs for bilateral breast, regional lymph nodes and OARs were manually contoured
• a convolutional neural network was then trained to perform the same task
• results were compared using Dice similarity coefficient, Hausdorff distance and qualitative scoring from 10 institutions
• inter-observer variability, delineation and DVH impact was assessed
• results showed good correlation between the manual and auto-contours for both OARs and CTVs with minimal dosimetric differences

Question 17

clinical applications of ML - dosimetry

Answer 17

• compared a rainforest model treatment plan with a human generated plan
• the outcome was 89% of ML plans were clinically acceptable and 72% were selected over human generated plans
• the ML plans showed a 60.1% reduction in median time for the entire RT planning process

Question 18

how will AI change clinical practice for RTs

Answer 18

• reduction of repetitive low value tasks (eg. contouring)
• increased ability to make complex decisions using decision aids
• more personalised treatments
• maintenance of AI and data systems

Question 19

what is the role of RTs, ROs and MPs as AI is rolled out

Answer 19

• multidisciplinary approach required
• benefits will vary depending on the application and between groups
  
  • decision aids most useful for ROs and RTs
  • quality assurance applications most useful for RT’s and MP’s
  • auto contouring beneficial for RT’s and RO’s
• tripartite discussion underway about the roles of each discipline in the context of AI and how to ensure safe and effective rollout

Question 20

how do we ensure safety of our patients is maintained

Answer 20

• relies on an understanding of how AI systems are designed and applied
• education of undergraduate and qualified staff of AI basics
• ability to understand the assumptions and limitations of an AI application to ensure accountability in decision making
• communication with patients about how these tools are used