rachitgoyell/dermoraSkinStressDataset

Dermora Skin Stress Dataset

A dermatology image dataset designed to study the relationship between psychological stress and visible skin conditions. Unlike standard skin disease datasets that focus purely on lesion classification, this dataset introduces a psychodermatology perspective — capturing not just what condition is present, but whether it is stress-triggered or stress-correlated.

This dataset is the training foundation for the Dermora project, an AI-assisted skin health monitoring system that combines dermatology and mental health signals.

---

Why This Dataset Exists

Most publicly available dermatology datasets — including HAM10000 — are built for disease classification in isolation. They do not account for the well-documented clinical relationship between psychological stress and skin condition onset or flare severity.

Conditions like acne, eczema, and psoriasis are routinely triggered or worsened by elevated stress. A dataset that treats these conditions as purely biological misses a critical dimension of the problem. This dataset is a step toward bridging that gap.

Existing Datasets	This Dataset
Classify skin lesions only	Classifies skin conditions with stress correlation
No mental health context	Designed for psychodermatology research
Disease-centric labeling	Stress-aware labeling
Not suitable for stress-skin modeling	Built specifically for stress-skin AI systems

---

Dataset Description

Domain

Dermatology + Psychodermatology (Mind-Skin Connection)

Modality

Images — skin condition photographs across facial and body regions

Conditions Covered

• Acne
• Eczema
• Psoriasis
• Normal / Healthy Skin

Labels

Skin condition labels:

• Acne
• Eczema
• Psoriasis
• Normal

Stress correlation labels (where annotated):

• Low Stress
• Medium Stress
• High Stress

---

Dataset Structure

dermoraSkinStressDataset/
├── train/
│   ├── acne/
│   ├── eczema/
│   ├── psoriasis/
│   └── normal/
├── validation/
│   ├── acne/
│   ├── eczema/
│   ├── psoriasis/
│   └── normal/
├── test/
│   ├── acne/
│   ├── eczema/
│   ├── psoriasis/
│   └── normal/
└── metadata.json

---

Intended Use Cases

• Skin condition classification from images
• Stress-level inference from visual skin biomarkers
• Psychodermatology research and modeling
• Fine-tuning vision foundation models (ViT, DINOv2, CLIP) on dermatology data
• Building stress-aware health monitoring applications
• Educational and prototyping use in AI + healthcare

---

How to Load

from datasets import load_dataset

dataset = load_dataset("rachitgoyell/dermoraSkinStressDataset")

# View a sample
print(dataset["train"][0])

Loading a Specific Split

train_data = load_dataset("rachitgoyell/dermoraSkinStressDataset", split="train")
val_data   = load_dataset("rachitgoyell/dermoraSkinStressDataset", split="validation")
test_data  = load_dataset("rachitgoyell/dermoraSkinStressDataset", split="test")

Using with a Vision Model

from datasets import load_dataset
from transformers import AutoImageProcessor
from PIL import Image

dataset = load_dataset("rachitgoyell/dermoraSkinStressDataset", split="train")
processor = AutoImageProcessor.from_pretrained("rachitgoyell/dermora-dinov2-vit-b-skin")

def preprocess(example):
    inputs = processor(images=example["image"], return_tensors="pt")
    return inputs

processed = dataset.map(preprocess, batched=True)

---

Limitations

• Lighting and image quality variance — images collected across different devices and conditions may introduce inconsistency
• Skin tone representation — the dataset may not be uniformly representative across all skin tones; performance of models trained on this data may vary across demographics
• Indirect stress inference — stress is not directly observable from skin images; stress labels are inferred or annotated based on clinical correlation, not measured physiological stress values
• Limited clinical validation — annotations were not produced under a formal clinical annotation protocol with dermatologist consensus
• Class imbalance — certain conditions may be over- or under-represented relative to real-world prevalence

---

Ethical Considerations

• This dataset is intended for research and educational use only
• It must not be used to build systems that autonomously diagnose or treat skin conditions
• Stress inference from visual data is indirect and should never be used as a standalone mental health assessment
• Models trained on this dataset should be deployed with human clinical oversight
• Care must be taken to evaluate and mitigate demographic bias before any downstream deployment

---

Related Resources

• Model trained on this dataset: rachitgoyell/dermora-dinov2-vit-b-skin
• Base model used: facebook/dinov2-base

---

License

Released under the MIT License.

This dataset is intended for research and prototyping use only. It is not intended for clinical deployment. Users are responsible for ensuring appropriate ethical review and regulatory compliance before using this data in any health-related application.