Niranjani Komethagavel
← Back to projects

Clarifeye

Smartphone-based fundus imaging + AI screening for retinal disease in LMICs.

Medical Imaging · Computer Vision · Global Health

Clarifeye — AI-Assisted Smartphone Fundus Screening

Clarifeye is a low-cost, smartphone-based fundus imaging system designed to enable scalable retinal disease screening in resource-constrained settings. The project combines a physical imaging solution with an AI-driven pipeline to support early detection of conditions such as diabetic retinopathy and glaucoma.

Rather than building a single model in isolation, the team approached Clarifeye as an end-to-end product: from image acquisition and usability, to data preprocessing, model readiness, and downstream clinical screening workflows.

My Role & Responsibilities

I served as team lead and AI engineer, owning both the technical direction and project execution.

Team leadership & execution

  • Organized and led weekly sprint-style meetings to set milestones, track progress, and unblock teammates
  • Allocated tasks across hardware, ML, and research tracks while maintaining alignment on the product vision
  • Helped teammates troubleshoot technical and conceptual hurdles, especially at the interface between imaging quality and ML performance

Product & concept development

  • Co-created the Clarifeye concept, including the smartphone-mount imaging approach and end-user workflow
  • Conducted early market and landscape research to understand gaps in existing retinal screening tools, particularly in low-resource settings
  • Framed the system around cost, portability, and reproducibility rather than lab-grade performance alone

AI & technical contributions

  • Co-designed the machine learning strategy with a strong emphasis on data quality, preprocessing, and evaluation readiness
  • Identified preprocessing as a critical bottleneck for medical imaging robustness and reproducibility
  • Built tooling to standardize and modularize image transformations across datasets and experiments

Technical Artifact: Medical Image Preprocessing Pipeline

One concrete output of my work on Clarifeye is a modular medical image preprocessing toolkit, built to support reproducible experimentation across medical image datasets.

This repository abstracts common preprocessing steps into plug-and-play transforms, making it easy to iterate on data preparation without rewriting pipelines or introducing experiment drift.

Key features

  • Modular transforms including:
    • Dark border cropping
    • Circular fundus cropping
    • CLAHE contrast enhancement
    • Resizing and sharpening
  • Centralized pipeline utilities for:
    • Run organization and naming
    • Consistent output directory structure
  • Jupyter-based workflow for:
    • Visual preview of transformations
    • Batch execution on train/test splits
  • Clear extension guide allowing new transforms to be added with minimal friction

Why this matters: In medical imaging, preprocessing decisions often dominate downstream model performance. This toolkit was designed to:

  • Make preprocessing explicit, inspectable, and reproducible
  • Enable rapid experimentation without code duplication
  • Serve as a foundation for downstream classification or segmentation models

Outcomes & Current Status

  • Established a scalable technical foundation for AI-assisted retinal screening
  • Delivered a reusable preprocessing framework applicable beyond Clarifeye
  • Aligned hardware, ML, and product considerations into a cohesive system design

While the project is early-stage, the emphasis on infrastructure, rigor, and reproducibility positions Clarifeye for future dataset expansion and model development.