Comparison of Artificial Intelligence based approaches to cell function prediction

Published

July 11, 2019

Author(s)

Sarala Padi, Petru S. Manescu, Nicholas Schaub, Nathan Hotaling, Carl G. Simon Jr., Peter Bajcsy

Abstract

Predicting Retinal Pigment Epithelium (RPE) cell functions in stem cell implants using non-invasive bright field microscopy imaging is a critical task for clinical deployment of stem cell therapies. Such cell function predictions can be carried out either by using Machine Learning (ML) or Deep Learning (DL) based approaches. These approaches have different tradeoffs in terms of prediction accuracies and their modeling complexities. This work aims at exploring the tradeoffs between ML and DL based approaches for RPE cell function prediction from microscopy images and understanding the accuracy relationship between pixel-, cell feature-, and implant label-level models. We compared accuracies of three ML and DL based approaches for cell function prediction and show that direct cell function prediction from images is more accurate than indirect DL and ML based predictions using intermediate segmentation and/or feature engineering steps. Furthermore, we evaluated accuracy variations with respect to model prediction configurations with and without transfer learning, and with five ML models and show that DL model outperform other models. Finally, we summarized the tradeoffs of the three approaches and quantified the linked accuracy relationships between segmentation and number of training samples, segmentation and cell features, and cell features and implant labels. We also showed that for the RPE cell data set, there is a clear relationship between the number of training samples, image segmentation accuracy, and cell feature histogram dissimilarity but there is no relationship between segmentation accuracy and the accuracy of RPE implant labels.

Citation

Elsevier

Volume

18, 2020

Issue

2352-9148

Pub Type

Journals

Download Paper

DOI Link

Keywords

cell segmentation, cell function prediction, comparison of machine learning and deep learning models, transfer learning

Citation

Padi, S. , Manescu, P. , Schaub, N. , Hotaling, N. , Simon, C. and Bajcsy, P. (2019), Comparison of Artificial Intelligence based approaches to cell function prediction, Elsevier, [online], https://doi.org/10.1016/j.imu.2019.100270 (Accessed December 30, 2024)