# Estimating Fibrosity Scores of Plant-Based Meat Products from Images: A Deep Neural Network Approach

## TL;DR

This paper introduces a deep neural network that estimates fibrosity scores of plant-based meat products from images, showing improved performance and explainability.

## Contribution

A novel deep neural network approach for estimating fibrosity scores from images, with improved individual subject modeling and explainable features.

## Key findings

- The network performed better when trained on individual subject scores, capturing nuanced perception aspects.
- The network's estimates were influenced only by relevant features like food matrices and air cells.
- Extraneous factors did not affect the network's fibrosity score estimates.

## Abstract

This paper proposes a deep neural network to estimate the fibrosities of plant-based meat product images. Images of varying fibrous microstructures were collected for this purpose, which were subject to spatial preprocessing and data enhancement. Their corresponding fibrosity scores were provided by two human experts. This data was used to train the network and to analyze its performance. Various statistical performance metrics were applied to evaluate the accuracy of the trained network’s estimated scores. It was found that the network performed significantly better when trained separately with fibrosity scores of each individual subject than with their combined scores, indicating that it was able to capture nuanced aspects of a subject’s perception. Another study was directed at explainability of the network’s estimates. Using standard software, a set of synthetic images of varying shapes and sizes were created as inputs to the network. Visual inspection of the output scores indicated that its estimates were influenced only by those features (i.e., food matrices and air cells) that were directly relevant to fibrosity, and not by extraneous factors.