Three Dimensional Root CT Segmentation using Multi-Resolution Encoder-Decoder Networks.

Mohammadreza Soltaninejad,Marcus Griffiths,Michael P Pound,Craig J Sturrock,Tony P Pridmore

doi:10.1109/tip.2020.2992893

Mohammadreza Soltaninejad, Marcus Griffiths + Show 3 more

Open Access

https://doi.org/10.1109/tip.2020.2992893

Copy DOI

Abstract

We address the complex problem of reliably segmenting root structure from soil in X-ray Computed Tomography (CT) images. We utilise a deep learning approach, and propose a state-of-the-art multi-resolution architecture based on encoderdecoders. While previous work in encoder-decoders implies the use of multiple resolutions simply by downsampling and upsampling images, we make this process explicit, with branches of the network tasked separately with obtaining local high-resolution segmentation, and wider low-resolution contextual information. The complete network is a memory efficient implementation that is still able to resolve small root detail in large volumetric images. We compare against a number of different encoder-decoder based architectures from the literature, as well as a popular existing image analysis tool designed for root CT segmentation. We show qualitatively and quantitatively that a multi-resolution approach offers substantial accuracy improvements over a both a small receptive field size in a deep network, or a larger receptive field in a shallower network. We then further improve performance using an incremental learning approach, in which failures in the original network are used to generate harder negative training examples. Our proposed method requires no user interaction, is fully automatic, and identifies large and fine root material throughout the whole volume.

Highlights

R OOT phenotyping is the process of characterising, objectively and quantitatively, the root systems of plants [1]
We evaluate our proposed network architecture against state-of-the-art networks in semantic segmentation, all adapted for volumetric segmentation
We evaluate our method on volumetric Computed Tomography (CT) images of intact wheat roots grown in soil captured at the University of Nottingham’s Hounsfield Facility

Summary

Introduction

R OOT phenotyping is the process of characterising, objectively and quantitatively, the root systems of plants [1] It offers valuable insight into the way root systems develop, react to environmental changes and other external stimuli, and interact with their natural soil environment. Traditional approaches have involved separating the soil from the root by washing, acquiring and analysing visible-light images [2] This approach can offer a fairly high-throughput solution, but during the process the root structure will likely be altered, and some finer roots will be lost in the washing process. It is common for naturally 3D root structures to be flattened and imaged using flatbed scanners, losing valuable architectural information. Other approaches have involved growing in translucent gel, or other artificial media, preserving root structure and allowing images to be

Methods

Results

Discussion

Conclusion