Contrastive learning-driven framework for neuron morphology classification

Abstract The Neuron morphology classification is a critical task in neuroscience research, as the morphological features of neurons are closely linked to the functional characteristics of neural circuits. However, traditional classification methods often struggle with the complexity and diversity of...

Full description

Saved in:
Bibliographic Details
Main Authors: Yikang Jiang, Hao Tian, Quanbing Zhang
Format: Article
Language:English
Published: Nature Portfolio 2025-07-01
Series:Scientific Reports
Subjects:
Online Access:https://doi.org/10.1038/s41598-025-11842-w
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849763410038751232
author Yikang Jiang
Hao Tian
Quanbing Zhang
author_facet Yikang Jiang
Hao Tian
Quanbing Zhang
author_sort Yikang Jiang
collection DOAJ
description Abstract The Neuron morphology classification is a critical task in neuroscience research, as the morphological features of neurons are closely linked to the functional characteristics of neural circuits. However, traditional classification methods often struggle with the complexity and diversity of neuronal morphologies. To address this, we propose PRT-net, a network architecture specifically designed for neuron morphology classification. By incorporating innovative data augmentation strategies and a contrastive learning framework, PRT-net effectively improves classification performance and model generalization. PRT-net leverages Complex Residual Structures and TreeLSTM to efficiently model the local features and global dependencies of neuron morphology. To address issues of data scarcity and imbalance, we designed a tailored data augmentation strategy that simulates diverse morphological variations, enhancing model robustness. Experiments conducted on three public datasets—BIL, JML, and ACT—demonstrate that PRT-net achieves classification accuracies of 78.45%, 67.11%, and 58.95%, respectively, significantly surpassing existing state-of-the-art methods. Notably, it achieves improvements of 2.9 and 3.3 percentage points on the JML and ACT datasets, respectively. Through the introduction of multiple evaluation metrics, we comprehensively analyze the classification and clustering performance of the model, validating its strong adaptability to complex data distributions. This study provides an efficient solution for neuron morphology classification, advancing research in this domain.
format Article
id doaj-art-7130c45eeb664c94b887318972f4a5e2
institution DOAJ
issn 2045-2322
language English
publishDate 2025-07-01
publisher Nature Portfolio
record_format Article
series Scientific Reports
spelling doaj-art-7130c45eeb664c94b887318972f4a5e22025-08-20T03:05:25ZengNature PortfolioScientific Reports2045-23222025-07-0115111710.1038/s41598-025-11842-wContrastive learning-driven framework for neuron morphology classificationYikang Jiang0Hao Tian1Quanbing Zhang2Institute of Electronic Information Engineering, Anhui UniversityInstitute of Electronic Information Engineering, Anhui UniversityInstitute of Electronic Information Engineering, Anhui UniversityAbstract The Neuron morphology classification is a critical task in neuroscience research, as the morphological features of neurons are closely linked to the functional characteristics of neural circuits. However, traditional classification methods often struggle with the complexity and diversity of neuronal morphologies. To address this, we propose PRT-net, a network architecture specifically designed for neuron morphology classification. By incorporating innovative data augmentation strategies and a contrastive learning framework, PRT-net effectively improves classification performance and model generalization. PRT-net leverages Complex Residual Structures and TreeLSTM to efficiently model the local features and global dependencies of neuron morphology. To address issues of data scarcity and imbalance, we designed a tailored data augmentation strategy that simulates diverse morphological variations, enhancing model robustness. Experiments conducted on three public datasets—BIL, JML, and ACT—demonstrate that PRT-net achieves classification accuracies of 78.45%, 67.11%, and 58.95%, respectively, significantly surpassing existing state-of-the-art methods. Notably, it achieves improvements of 2.9 and 3.3 percentage points on the JML and ACT datasets, respectively. Through the introduction of multiple evaluation metrics, we comprehensively analyze the classification and clustering performance of the model, validating its strong adaptability to complex data distributions. This study provides an efficient solution for neuron morphology classification, advancing research in this domain.https://doi.org/10.1038/s41598-025-11842-wNeuron morphology classificationData augmentationContrastive learningTreeLSTM
spellingShingle Yikang Jiang
Hao Tian
Quanbing Zhang
Contrastive learning-driven framework for neuron morphology classification
Scientific Reports
Neuron morphology classification
Data augmentation
Contrastive learning
TreeLSTM
title Contrastive learning-driven framework for neuron morphology classification
title_full Contrastive learning-driven framework for neuron morphology classification
title_fullStr Contrastive learning-driven framework for neuron morphology classification
title_full_unstemmed Contrastive learning-driven framework for neuron morphology classification
title_short Contrastive learning-driven framework for neuron morphology classification
title_sort contrastive learning driven framework for neuron morphology classification
topic Neuron morphology classification
Data augmentation
Contrastive learning
TreeLSTM
url https://doi.org/10.1038/s41598-025-11842-w
work_keys_str_mv AT yikangjiang contrastivelearningdrivenframeworkforneuronmorphologyclassification
AT haotian contrastivelearningdrivenframeworkforneuronmorphologyclassification
AT quanbingzhang contrastivelearningdrivenframeworkforneuronmorphologyclassification