Related Experiment Video
Updated: Sep 2, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.9K
A deep learning-based algorithm for tall cell detection in papillary thyroid carcinoma
Sebastian Stenman1,2,3, Nina Linder1,4, Mikael Lundin1
1Institute for Molecular Medicine Finland - FIMM, University of Helsinki, Helsinki, Finland.
Plos One
|August 9, 2022
Summary
A new deep learning algorithm accurately detects tall cells (TCs) in papillary thyroid carcinoma (PTC). Higher TC scores identified by the algorithm predict poorer relapse-free survival in PTC patients.
Area of Science:
- Oncology
- Pathology
- Artificial Intelligence in Medicine
Background:
- The tall cell variant (TCV) of papillary thyroid carcinoma (PTC) is an aggressive subtype, but its definition leads to diagnostic challenges and interobserver variability.
- Accurate identification and quantification of tall cells (TCs) are crucial for assessing PTC aggressiveness and patient prognosis.
- Current methods for TC assessment are subjective, highlighting the need for objective and reproducible diagnostic tools.
Purpose of the Study:
- To develop and validate a deep learning algorithm for the automated detection and quantification of tall cells (TCs) in papillary thyroid carcinoma (PTC).
- To evaluate the algorithm's performance in comparison to human assessment.
- To determine the association between algorithm-derived TC scores and patient disease outcomes, specifically relapse-free survival.
Main Methods:
- A deep learning algorithm was trained using supervised learning on PTC samples.
- The algorithm was tested on an independent dataset and validated on 90 PTC samples from Helsinki.
- Algorithm-based TC percentages were compared to visual scoring by a human investigator and correlated with disease outcomes and tumor relapse samples.
Main Results:
- The deep learning algorithm achieved high sensitivity (93.7%) and specificity (94.5%) in detecting TCs.
- Algorithm-derived TC scores significantly correlated with diminished relapse-free survival at various cutoff points (10%, 20%, 30%).
- Visually assessed TC scores did not significantly predict survival, unlike the algorithm's findings.
Conclusions:
- A novel deep learning algorithm effectively detects tall cells in PTC, offering an objective measure.
- A high deep learning-based TC score is a statistically significant predictor of less favorable relapse-free survival in PTC.
- This AI-driven approach has the potential to improve the accuracy and consistency of PTC subtyping and prognostic assessment.
