Table 1 Comparison between traditional ophthalmic techniques and Embodied Artificial Intelligence (EAI) ophthalmic techniques

Data Source

Passive recording (manual examination, input by ophthalmologists)

Active multimodal sensing (intraoperative OCT, eye tracking, etc.)

Information Integration

Unimodal or limited multimodal (e.g., OCT and CFP)

Multimodal fusion (eye tracking, biomechanics, etc)

Feedback Mechanism

Doctor explains results, patient passively receives

Device dynamically adjusts exams, interactive with users

Learning Mechanism

One-way learning, requires large scale datasets for retraining

Closed-loop perception-action reinforcement learning

Decision Timeliness

Retrospective analysis (minutes)

Real-time feedback (milliseconds)

Decision Capability

Relies on subjective judgment (affected by fatigue/experience)

Instant data acquisition and feedback

Adaptability

Limited adaptability to novel or rare diseases.

Requires extensive training and sensor integration for practical deployment.

Cost

Relatively low for standard techniques (e.g., slit-lamp examination).

Relatively high, but it can reduce human workload in long-term use through automation.

Clinical application

Standardized screening and diagnosis (e.g., DR grading, OCT analysis)

Personalized treatment assistance (e.g., retinal injections, robotic obstacle avoidance)

Pros and Cons

Well-established, cost-effective, but limited real-time adaptability.

Enables real-time, data-driven decision-making, but requires high computational resources and advanced integration.