Semantic Segmentation of Neuronal Bodies in Fluorescence Microscopy Using a 2D+3D CNN Training Strategy with Sparsely Annotated Data

TL;DR

This paper introduces a two-phase training strategy for 3D CNNs that leverages sparse 2D annotations and a 2D CNN to improve neuronal structure segmentation in fluorescence microscopy, reducing annotation effort and training time.

Contribution

It presents a novel semi-supervised training approach that infers missing labels for 3D CNNs using a 2D CNN, enabling effective learning from sparse annotations.

Findings

Improved segmentation accuracy with sparse annotations.

Reduced manual annotation effort and training time.

Effective combination of 2D and 3D CNNs for neuronal imaging.

Abstract

Semantic segmentation of neuronal structures in 3D high-resolution fluorescence microscopy imaging of the human brain cortex can take advantage of bidimensional CNNs, which yield good results in neuron localization but lead to inaccurate surface reconstruction. 3D CNNs, on the other hand, would require manually annotated volumetric data on a large scale and hence considerable human effort. Semi-supervised alternative strategies which make use only of sparse annotations suffer from longer training times and achieved models tend to have increased capacity compared to 2D CNNs, needing more ground truth data to attain similar results. To overcome these issues we propose a two-phase strategy for training native 3D CNN models on sparse 2D annotations where missing labels are inferred by a 2D CNN model and combined with manual annotations in a weighted manner during loss calculation.