Potential for AI in clinical optometry practice and education

Key Points
  • Artificial intelligence (AI) is intelligence demonstrated by computers as opposed to intelligence demonstrated by humans.
  • Machine learning describes the manner in which the computer applies AI.
  • Algorithms are sets of instructions that a computer may follow, they do not necessarily require machine learning, such as a flow chart for use in managing a virtual patient.
  • An algorithm that does use machine learning is the application of Bayes theorem and, with greater complexity, the Google Mind/Moorfields/UCL interpretation of OCT data.
  • Bayes theorem involves updating existing beliefs with new data to modify a belief.
  • Application to eye disease would first involve establishing initial beliefs – how common one or more outcomes are. Examples of outcomes might include specific eye conditions or urgency of referral. Pre-test odds may be derived for these initial beliefs.
  • Objective new information may include presenting history, signs and test result data. Diagnostic value of these may then be expressed in terms of likelihood ratios which can be used to improve efficiency of assessment by identifying which have the greatest diagnostic value.
  • The improved belief is expressed as post-test odds and is calculated by multiplying the pre-test odds by the likelihood ratio. Post-test odds can be used to compare the relative benefit of alternative outcomes.
  • Storage of personal data, as required for machine learning, can easily be erased after interpretation and so data protection rules are easily followed
  • SITA used in fields testing employs Bayes theorem to shorten testing times for patients with suspect glaucoma.
  • Use of deep learning architecture in OCT data analysis has showed 94% accuracy in recommending correct referral of patients, matching or exceeding the referral accuracy of specialist clinicians.
  • Innovative clinicians should plan for a future where artificial intelligence plays a greater role in practice.
  • Advantages in AI application in eye care include better patient care, improved medico legal protection for practitioners, fingertip access to all patient data to aids support challenging management decisions and greater resource for both legal and educational bodies in optometry.
  • AI has potential to allow greater time for practitioners to focus on the practitioner and patient relationship.
  • Clarity of the processes used by algorithms may reassure clinicians regarding their decision making.
  • The Bayes translational learning cycle (BTLC) will enable optometrists of the future to be involved in AI and its use in clinical optometry and might easily be incorporated into a three-year training programme to enhance clinical understanding in students.