Power Hour Optometry

Episode Summary: Artificial Intelligence in Eye Care – What's Real, What's Overhyped & How to Evaluate It

Host: Eugene Shatsman | Guest: Dr. Abed Sarhan
Date: April 10, 2026

Episode Overview

This episode of Power Hour dives deep into the real-world application of artificial intelligence (AI) in optometry. Eugene Shatsman hosts Dr. Abed Sarhan, a PhD in deep learning with experience spanning over a decade before AI became mainstream. Together, they cut through marketing hype to discuss which AI tools are delivering tangible value in clinical eye care today, what remains overpromised, and, crucially, how practice owners should evaluate AI solutions for their workflows and patients. The conversation also explores where AI may transform the future of eye care—and where the irreplaceable human elements remain.

Key Discussion Points & Insights

1. Defining AI, Machine Learning, and Deep Learning

(02:32–06:01)

• Dr. Sarhan explains the difference:
  • Machine Learning: "Machine learning is an umbrella... You're trying to build a model and you feed it with 10 variables, and the model will output if this person has a condition or not." (03:59)
  • Deep Learning: "Deep learning goes way ahead. We’re talking about 10,000 variables... fed into a series of networks that get filtered... and then give you an output." (03:59)
• Practical vs. Academic AI: Dr. Sarhan shares that AI "has been there for decades," but practical and widespread use in clinics was triggered by advancements in computing hardware (TPUs), which enabled high-precision analysis of vast medical datasets. (06:43–08:01)

2. Where AI Excels—And What’s Still Missing

(09:13–15:08)

• Progress vs. Hype:
  • "I think 11 years ago people were overpromising what AI... can deliver. I don't think we hit that. I think there's so much hype about AI. We are still in the middle of... good evaluation." — Dr. Sarhan (09:13)
• Misunderstandings in Evaluation:
  • Many in healthcare use “accuracy” and “AUC” metrics in unspecialized ways. Not all metrics reflect real-world performance, especially for nuanced tasks like image segmentation or voice agents.
  • "Many people miss asking the right question for the right tool." (10:20)
  • Example: Segmentation should be evaluated by border detection, not just overall accuracy if the structure occupies most of the image. (11:11)
  • Voice tools must also be tested for emotional tone, accent, and stress—not just word error rate. (13:29)

3. Gaps in AI for Eye Care

(15:08–19:20)

• Trust and Transparency:
  • Critical gaps include the lack of clear, explainable decision-making in AI ("black box diagnostics"), and insufficient validation or transparency from vendors.
  • "The gap in the field now is... how you provide explanation to the doctor about specific decision that's been made by the AI." (16:49)
• What’s Needed: High-precision segmentation, explainability, and continual evolution/adaptation of tools are essential for trust and clinical adoption.

4. Where AI Will Transform Eye Care in the Near Future

(17:58–23:53)

• Most Impactful Areas:

  • Examinations/Prescreening: "Very dependent on staff and prone for friction... AI can transform how exams are delivered." — Dr. Sarhan (18:04)
  • Treatment Selection & Explanation: AI can support doctors in effectively explaining consequences and benefits to patients, potentially simulating the impact of treatment vs. non-treatment. (19:20–23:53)

• Balance of Tech & Human Touch:

  • While AI can simulate scenarios, the doctor’s relationship, empathy, and trust remain essential.
  • "The doctor will be able to decide... and have a conversation, but the patient will have the capabilities of seeing [the outcomes] in real life or virtually." (21:58)

5. Evaluating Current and Future AI Tools

(26:57–44:06)

• AI Technologies to Adopt Now:

  • Voice Agents: Offload repetitive front-desk tasks, such as appointment scheduling and basic queries—reduced unanswered calls and improved patient access.
  • Patient Education Tools: Personalized AI-driven education to enhance treatment understanding/compliance.
  • OCT/Image Analysis Tools: Consider newer tools, but understand their limitations and evidence.
  • Data Integration Systems: Tools that unify fragmented device outputs and patient data.

• Key Adoption Advice:

  • Implement incrementally; start with niche, evolving tools.
  • Test solutions thoroughly—including in-house stress-testing with staff.
  • Look for tools where the vendor clearly communicates capabilities, metrics (broken down by variable), and limitations.

"Once you know the limitations, you don't really need the voice agent to replace everything... you can integrate it in a staged way to actually complement what you have." — Dr. Sarhan (36:36)

• Maintain the Human Element:
  • AI can automate routine tasks, but humans remain vital for specialization, managing complexity, and building relationships.

6. Overhyped Promises and Cautions

(49:54–54:42)

• Beware of Bold Claims:

  • "People who's telling you we can detect glaucoma... or diabetic retinopathy. I think this is overhyped." (50:37)
  • Be wary of “black box” diagnostics; insist on tools that explain how decisions are made, with solid research and validation.

• Critical Evaluation Questions:

  • For any AI solution, ask:
    • How is the decision made? Is there transparency?
    • What specific clinical metrics are reported and validated?
    • How does it handle data from different manufacturers and formats?
    • What support and evolution capacity does the vendor/team have?
    • Is there evidence of benefit in real clinical settings?

7. Automation, Efficiency & Workflow: Where AI Can Help

(55:27–58:32)

• Front Desk/Admin Automation: Staff freed from repetitive work, able to focus on higher-value activities.
• Pre-Screening & Data Integration: Portable, automated devices can streamline exams; better data flow reduces bottlenecks.
• EHR Integration is Crucial: Central data hub for practice—solutions must integrate effectively.
• Always Prioritize Patient Experience:
  • "If it's not inherently designed with the patient in mind first, then... it's a failure of design." — Eugene Shatsman (57:31)

8. The Future of Eye Care & Final Thoughts

(44:50–61:05)

• More Digitization, Less Routine Labor:
  • Clinics will become "more digitized, more connected, with less staff required but greater patient impact." (45:42)
  • However, universal device/software integration is not reality yet; clinics using mixed equipment face hurdles.
• Vendors Often Miss the Mark:
  • "In many situations they don't really speak with the customer... it's important to differentiate: building great machines doesn't necessarily mean building great software." — Dr. Sarhan (48:45)
• Advice for Practice Owners:
  • Start integrating AI now, but do it thoughtfully.
  • "Machine learning... it's not going to be for replacing doctors. It's going to be for assisting them in making better decisions... improving their workflows." (59:32)
  • Always ask hard questions; don’t go blind into AI investment.

Notable Quotes & Memorable Moments

• On AI hype vs. reality:

  "I think there's so much hype about AI. We are still in the middle of... a good evaluation." — Dr. Abed Sarhan (09:13)

• On transparency in AI tools:

  "AI modules that are black boxes in my opinion need to be avoided." (52:16)

• On the ongoing role of doctors:

  "AI will not replace the relationship building between the doctor and the patient." (37:57)

• On solving real clinical problems:

  "If it's not inherently designed with the patient in mind first, then... it's a failure of design." — Eugene (57:31)

Important Timestamps

• Defining AI, ML, and Deep Learning: 02:32–06:01
• Hype vs. Reality & Evaluation Gaps: 09:13–13:29
• Impactful Future Areas for AI in Eye Care: 17:58–23:53
• Practical Evaluation Tips for AI Tools: 26:57–36:36
• Cautions & Overhyped Claims: 49:54–54:42
• Final Thoughts on Integration & the Human Element: 59:32–61:05

TL;DR Key Takeaways

• AI has real, practical benefits in operational efficiency and patient engagement—but not all promises are rooted in reality.
• Evaluate each technology on its specialized metrics, real clinical evidence, and transparency—avoid black box solutions.
• Start integrating AI thoughtfully now to keep your practice competitive and patient-focused.
• The human element—especially in specialization and relationship—will always have a role that AI can't replace.