Artificial neural networks for 3D cell shape recognition from confocal images

TL;DR

This paper introduces a dual-stage neural network system that analyzes 3D cell shapes from microscopy images, using spherical harmonics for feature extraction to enable accurate, unbiased classification for medical diagnostics.

Contribution

The novel dual-stage neural network architecture effectively captures fine 3D shape details and automates cell classification based on spherical harmonics features, improving diagnostic accuracy.

Findings

Successfully tested on red blood cells from healthy donors and patients

Revealed characteristic shape features through spherical harmonics spectrum

Provides reproducible and unbiased shape recognition for diagnostics

Abstract

We present a dual-stage neural network architecture for analyzing fine shape details from microscopy recordings in 3D. The system, tested on red blood cells, uses training data from both healthy donors and patients with a congenital blood disease. Characteristic shape features are revealed from the spherical harmonics spectrum of each cell and are automatically processed to create a reproducible and unbiased shape recognition and classification for diagnostic and theragnostic use.