Dermal collagen fiber data for Fiber Classification by ResNet

Type I collagen is a major component of the dermis and is formed by dermal fibroblasts. The development of dermal collagen structures has not been fully elucidated despite the major presence and importance of the dermis. This Lack of understanding is due in part to the opacity of mammalian skin and it has been an obstacle to cosmetic and medical developments. We succeeded in revealing the process of dermal collagen formation using the highly transparent skin of the axolotl and fluorescent collagen probes. We clarified that epidermal cells, not dermal fibroblasts, primarily contribute to dermal collagen formation. Fibroblasts play a role in modifying the collagen fibers already built by keratinocytes. We confirmed that collagen production by keratinocytes is a widely conserved mechanism in other model organisms. Our findings demand the current consensus about dermal collagen formation to change and lead to innovations in cosmetology and skin medication., For dimension reduction of collagen image data, we used 484 collagen images from the confocal images in axolotl dorsal skin for the analysis and adopted ResNet50 with ImageNet-pretrained weights (ResNet50_Weights.IMAGENET1K_V2 from the torchvision library) for feature extraction. We extracted image features from the last global average pooling layer. Then, linear PCA was applied to the 2048-dimensional feature vectors. We performed no training of the neural network with our data; hence, the feature extraction was agnostic on which developmental stages the images came from. , , # Dermal collagen fiber data for Fiber Classification by ResNet

[https://doi.org/10.5061/dryad.6q573n66z](https://doi.org/10.5061/dryad.6q573n66z)

## Description of the data and file structure

#### File: Ohashi\_et\_al\_2024.zip

**Description:** Collagen fibers in the axolotl dermis

The folder names (\"5cm, 8cm, 10cm, 12cm\") indicate the animal sizes.
Each file has name `#a-bb-c`

* \"a\" for a sample name,
* \"b\" for a slice name, and
* \"c\" for a position name of an identical slice.

All files are saved as a jpeg (jpg) file.

In the 10cm and 12cm folders, there are subfolders, entitled the sample name. The sample names are consistent with the names in the paper published in Nature Communication in 2025, Ohashi et al., Figure 9. 

Those images were used for ResNet analysis in the paper by Ohashi et al.. 2025.

## Code/software

N/A

## Access information

Other publicly accessible locations of the data:

* n/a

Data was derived from the following sources:

* n/a

I型胶原蛋白（Type I collagen）是真皮的主要组成成分，由真皮成纤维细胞合成。尽管真皮在机体中占比颇高且功能关键，但其胶原结构的发育机制尚未完全阐明。这种认知空白部分源于哺乳动物皮肤的不透性，这也成为了美容与医学领域发展的阻碍。本研究借助美西螈高度透明的皮肤与荧光胶原探针，成功揭示了真皮胶原的形成过程。我们明确了，主要参与真皮胶原形成的并非真皮成纤维细胞，而是表皮细胞；成纤维细胞的作用是对角质形成细胞已构建的胶原纤维进行修饰。我们还证实，角质形成细胞产生胶原的机制在其他模式生物中广泛保守。本研究的发现要求学界更新当前关于真皮胶原形成的主流共识，并有望推动美容学与皮肤药物学领域的创新。

为实现胶原图像数据的维度缩减，本研究使用了美西螈背部皮肤共聚焦成像（confocal images）中的484张胶原图像用于分析，并采用搭载ImageNet预训练权重的ResNet50（torchvision库中的ResNet50_Weights.IMAGENET1K_V2）进行特征提取。我们从最后一层全局平均池化层提取图像特征，随后对2048维的特征向量应用线性主成分分析（Principal Component Analysis, PCA）。本研究未使用自有数据对神经网络进行训练，因此特征提取过程与图像所属的发育阶段无关。

# 用于ResNet纤维分类的真皮胶原纤维数据集
https://doi.org/10.5061/dryad.6q573n66z

## 数据与文件结构说明

#### 文件：Ohashi_et_al_2024.zip
**描述**：美西螈真皮中的胶原纤维
文件夹命名格式为“5cm、8cm、10cm、12cm”，代表受试动物的体长。
每个文件的命名格式为`#a-bb-c`：
- `a` 代表样本名称，
- `bb` 代表切片编号，
- `c` 代表同一切片内的位置编号。
所有文件均保存为JPEG（jpg）格式。
在10cm与12cm的文件夹中，存在以样本名称命名的子文件夹，其样本名称与2025年发表于《Nature Communications》的Ohashi等人论文中的图9所用名称一致。
上述图像被用于Ohashi等人2025年论文中的ResNet分析。

## 代码/软件
无

## 获取信息
该数据的其他公开获取渠道：
无
本数据衍生自以下来源：
无