Skip to content

Roadmap

This roadmap outlines the major phases of the Global Oculomics Initiative. Each phase builds toward the central objective of creating the first multimodal foundational model for oculoplastic and craniofacial disease.

The project proceeds in five coordinated phases:

Roadmap Overview

See below for details on each specific phase

Phase 1 — Dataset Assembly

Dataset development is the rate-limiting step for all downstream work. Two datasets must be completed before pretraining begins.

Focus Areas

  • Foundational Model pretraining dataset (open-source)
  • UIC retrospective clinical dataset

Key Outputs

  • Cleaned, large-scale pretraining dataset
  • PD extraction at scale
  • Multimodal dataset index

Status: In Progress

Phase 2 — Global Deployments

Global data collection requires long lead times due to IRB processes, site logistics, and training. This phase runs in parallel with Phase 1.

Goals

  • Deploy Glorbit across 3–4 international sites
  • Assemble prospective, globally diverse datasets
  • Enable cross-site harmonization and external validation

Status: In Progress

Phase 3 — Multimodal Pretraining

Pretraining involves learning a multimodal backbone that binds image embeddings and periorbital distance features.

Dependencies

  • Completed pretraining dataset
  • Validated PD extraction pipeline

Key Outputs

  • Multimodal embedding space
  • Ablation studies
  • Pretraining technical report

Status: Begins after Phase 1 completion

Phase 4 — Fine-Tuning on Clinical and Functional Tasks

After pretraining, the backbone is adapted to downstream oculoplastics and craniofacial tasks.

Downstream Tasks

  • Multiclass disease classification
  • Craniofacial disorder classification
  • GVF regression/classification
  • Ptosis outcomes modeling
  • Lesion / eyelid pathology classification

Status: Begins after Phase 3

Phase 5 — Validation, Deployment, and Manuscripts

This phase focuses on internal/external validation, clinical interpretation, and manuscript writing.

Internal Validation

  • UIC datasets
  • Craniofacial datasets
  • Longitudinal subsets

External Validation

  • Global deployments
  • Partner hospitals (Mt. Sinai, others)

Key Outputs

Status: Ongoing (late project)