# Adapting a machine learning algorithm to predict thyroid cytopathologyin LMIC

> **NIH NIH R21** · DUKE UNIVERSITY · 2022 · $187,615

## Abstract

Abstract
Pathology expertise and services in low and middle income countries (LMIC) are severely inadequate and
limited. One area in which pathology is especially critical for diagnosis and clinical decision making is thyroid
related disease and cancers. The incidences of thyroid nodules and cancers have increased worldwide. A
fundamental part of the clinical assessment for thyroid nodules is ultrasound-guided fine needle aspiration biopsy
(FNAB). Unfortunately, this cytopathology expertise is scarce in LMIC and many nodules are surgically removed
for diagnosis. Since 90-95% of these nodules are without malignancy, many patients undergo unnecessary
surgery with the associated risks, financial costs, and stressing already limited resources. What is critically
needed is a means to provide cytopathology expertise to interpret thyroid FNAB in LMIC.
The overall goal of this proposal is to implement our ML approach into low resource settings to provide
accurate and timely cytopathology analysis of FNAB specimens. This R21/33 proposal focuses on MLA
adaptation using smartphones for LMIC (R21) and subsequent capacity building and implementation (R33)
through well-established LMIC research partnerships in tertiary centers in Tanzania and Vietnam. We
hypothesize that our MLA can use smartphone captured images to assess probability of malignancy in thyroid
FNAB that is comparable to trained cytopathologists.
Our Aims are: R21: Adaptation of a smartphone approach to capture cytopathology images for MLA analysis.
Archived FNAB samples used at different international institutions will be used to train and adapt the MLA with
these images. Furthermore, we would train local personnel in LMIC to image capture using this new setup. The
results from this phase will be an adapted MLA using smartphone images that has comparable sensitivity and
specificity to the original MLA. R33: Implementation of the smartphone MLA approach into LMIC clinical settings.
A prospective implementation of this approach in tertiary centers in Tanzania and Vietnam will be conducted.
FNAB samples obtained would undergo standard of care pathology assessment as well as be analyzed with the
MLA protocol. Local cytopathologist assessment will be compared with the MLA analysis. Outcome measures
include MLA assessment of slides, technical slide review and image capture, concordance with pathology
between MLA and expertise both in US and LMIC.
The expected outcome will lead to a cost-effective method to implement MLA for thyroid FNAB that can be used
in LMIC. This approach will not only assist in early diagnosis of thyroid cancers but also improve the utilization
of limited resources through effectively identifying those that need surgery from those that do not.

## Key facts

- **NIH application ID:** 10458057
- **Project number:** 5R21CA268428-02
- **Recipient organization:** DUKE UNIVERSITY
- **Principal Investigator:** WALTER Tsong LEE
- **Activity code:** R21 (R01, R21, SBIR, etc.)
- **Funding institute:** NIH
- **Fiscal year:** 2022
- **Award amount:** $187,615
- **Award type:** 5
- **Project period:** 2021-08-01 → 2024-07-31

## Primary source

NIH RePORTER: https://reporter.nih.gov/project-details/10458057

## Citation

> US National Institutes of Health, RePORTER application 10458057, Adapting a machine learning algorithm to predict thyroid cytopathologyin LMIC (5R21CA268428-02). Retrieved via AI Analytics 2026-05-23 from https://api.ai-analytics.org/grant/nih/10458057. Licensed CC0.

---

*[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*