Dermal collagen fiber data for Fiber Classification by ResNet
Data files
Jan 28, 2025 version files 4.07 MB
-
Ohashi_et_al_2024.zip
4.07 MB
-
README.md
981 B
Abstract
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.
README: Dermal collagen fiber data for Fiber Classification by ResNet
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
Methods
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.