Evaluating the Accuracy of Traction Force Estimations in Traction Microscopy via Deep Learning

Author

Sunimali Jayamini Rathnayake, Southern Illinois University CarbondaleFollow

Date of Award

8-1-2025

Degree Name

Master of Science

Department

Electrical and Computer Engineering

First Advisor

Tragoudas, Spyros

Abstract

This research introduces a deep learning-based framework for evaluating the structural consistency of traction force estimations in microscopy. The proposed approach identifies deviations in predicted traction maps by comparing their boundary structures to those from a chosen baseline method, without requiring ground truth labels.A UNET-based model is trained using both clean and synthetically distorted data, incorporating a set of noise models that simulate biologically relevant perturbations. This training strategy improves the model's ability to flag potentially unreliable traction patterns across varying distortion levels.To quantify these deviations, we introduce a metric ranging from 0 to 1, allowing for threshold-based identification of inconsistent traction outputs. The framework generalizes across multiple cell samples cultured on the same substrate and supports automated, scalable evaluation without manual inspection. Results demonstrate the model’s robustness under diverse scenarios and its practical value in replacing time-consuming manual evaluation with an automated and scalable assessment framework.