Directions Brain

2026-05-19-saar-leshno-glaucoma-data-tracking

5 min read

Saar/Guy ↔ Ari Leshno — glaucoma visual-field tracking pitch

Source: raw/meetings/Saar<>Leshno (2026-05-19 18.00).txt

Language note: Hebrew ASR transcript, ~62 minutes. The opening ~6 minutes are off-topic chitchat (Maccabi/football, the Karpathy-to-Anthropic rumor, a Google-Meet vs Zoom rant) and are excluded from this summary. Ari’s name is rendered “Ari Leshno” in the transcript’s speaker labels — earlier ASR transcripts had captured this same person as “הארי”; the prior hari page was retired and merged into ari-leshno alongside this ingest (see refactor entry in log.md).

Summary

saar-arbel and guy-barkat met ari-leshno (introduced by Ohad) to scope a healthcare opportunity. Ari is an ophthalmologist + glaucoma specialist at sheba and head of the eye institute’s research-and-innovation unit; he has a Sheba budget + green light to fund a startup and is actively recruiting entrepreneurs. He pitched a concrete first wedge — OCR-based cross-clinic ingest of visual-field test PDFs (Humphrey perimetry / HFA), competing with Zeiss’s Forum software (~$100K incumbent) — and walked through the bigger ophthalmology-AI roadmap behind it. Saar/Guy committed to deciding “in the next few days” whether to take the partnership; as of the 2026-05-27-directions-vertical-pivot-and-prediction-markets sync 8 days later, no decision was logged.

Key takeaways

  • The wedge: glaucoma visual-field tracking, OCR-first. Visual-field tests have not changed in 40 years; clinicians today literally print PDFs and visually compare them across years to spot deterioration. Zeiss’s Forum software (~100/clinic/month subscription. Sizing Ari gave: ~35K ophthalmology clinics globally × 10% × $100/mo ≈ “tens of millions per year.” Explicitly framed as “not a unicorn but a real wedge into the eye-care vertical.” See clinical-data-portability.
  • Sheba is offering capital + green light + raw-data access. Ari and yael have an internal Sheba budget to develop the product and authority to engage Sheba’s AI center for OCR development. They also hold the raw underlying data (decades of Excel exports + scanned PDFs from Sheba’s archive). Ari is shopping the project to entrepreneurs — they spoke to “someone last week” before this call — and wants Saar/Guy back with a yes/no within days.
  • Underlying generalizable pattern: chronic-disease longitudinal data tracking. Ari volunteered the analogy to diabetes and hypertension: any chronic disease where multi-year trends drive the treatment decision suffers the same data-fragmentation problem. The structural cause is that instrument vendors (e.g. zeiss) have no incentive to support cross-vendor import — proprietary formats are part of the lock-in that drives instrument sales. DICOM is theoretically standard but isn’t interoperable in practice. Captured as clinical-data-portability.
  • US is behind Israel on EHRs by ~a decade. Some US hospitals/clinics still use paper and pen. Israeli ophthalmology runs on kameleon (which Ari called “a junk monopoly”); the US niche-EHR for ophthalmology he named was “Apik / Epic” (likely a niche product, not Epic Systems the giant). Implication: the US is the larger TAM but the integration baseline is lower, which actually favors a PDF-OCR product over a deep-integration play.
  • Two explicit “don’ts” from Ari. (1) Don’t enter the AI scribe / note-taker space — saturated, “many people are trying.” (2) Don’t build diagnostic AI — every diagnosis-as-output crosses an FDA gate. The workaround precedent is idx / “AI-Doc”: phrase output as “this scan is normal” (a screening assertion) rather than a diagnosis, and FDA clearance becomes tractable. The bigger plays — deterioration prediction, treatment recommendation, missed-stroke detection on retinal imaging — are real but explicitly the step after the visual-field-tracking wedge.
  • Why the wedge is competition-resistant. zeiss has no incentive to add cross-vendor PDF import to Forum (it would dilute their hardware lock-in). Comprehensive ophthalmologists (“the bulk of the market”) can’t afford a 100/mo product that meaningfully reduces visit time wins on price + ergonomics, not on AI smarts. sheba itself is developing comparable internal tooling, which is a competitive concern but also a validation signal.

Decisions

  • (no firm decisions — exploratory pitch. Saar/Guy committed to come back with a yes/no within “the next few days” but did not commit on the call.)

Action items

Open questions

  • Does the visual-field-tracking wedge actually generalize to a Brain-shaped product, or is it a thin SaaS that the Brain layer adds nothing to? If the latter, this is a healthcare startup, not a Brain vertical.
  • If we take the partnership, what is the equity / IP arrangement with sheba and with ari-leshno + yael? Not discussed on the call.
  • The “sheba is also developing this internally” line is unsettling — what does the IP / assignment look like, and how soon would internal Sheba dev catch up to a 30-min Claude MVP?
  • Does the FDA-gated long-term roadmap (deterioration prediction, etc.) require us to commit now to a regulatory path, or can the first 12 months be pure SaaS? The pitch implies the latter but it isn’t certain.
  • Is the chronic-disease-longitudinal-tracking generalization (clinical-data-portability) a one-vertical idea or a platform thesis? Need a concrete second vertical (diabetes? hypertension?) to test the pattern.

Graph View

Backlinks