Miguel764/efficientnetv2s-skin-cancer-classifier

EfficientNetV2S - Skin Cancer Classifier

This repository hosts the EfficientNetV2S deep learning model trained for skin cancer classification using the HAM10000 dataset. More information on github: Skin Cancer XAI - Universidad

The model is used in the SkinAI Diagnostics application — a FastAPI-based web tool that allows users to upload dermatoscopic images and receive predictions across 7 skin lesion categories.

🔗 Application GitHub Repository: Skin Cancer Classifier Fastapi

---

Model Overview

• Architecture: EfficientNetV2S
• Parameters: ~20.5 million
• Dataset: HAM10000
• Number of Classes: 7
  • Actinic Keratoses and Intraepithelial Carcinoma (AKIEC)
  • Basal Cell Carcinoma (BCC)
  • Benign Keratosis-like Lesions (BKL)
  • Dermatofibroma (DF)
  • Melanoma (MEL)
  • Melanocytic Nevi (NV)
  • Vascular Lesions (VASC)
• Calibration: Temperature Scaling (T = 2.77) to improve probability reliability
• Performance:
  • Accuracy: 0.88
  • Macro F1-score: 0.80
  • Expected Calibration Error (ECE): 0.022 (after T-scaling)

---

Usage in SkinAI Diagnostics

In the SkinAI Diagnostics app, the model file is automatically downloaded from Hugging Face if not found locally.
This ensures the application works without requiring the model to be stored in the GitHub repository.

Automatic download logic:

• Checks if app/model/efficientnetv2s.h5 exists locally.
• If not found, downloads from Hugging Face using huggingface_hub.
• Loads the model into TensorFlow/Keras for inference.

---

Direct Download

You can manually download the model by clicking the Files and versions tab above or using the huggingface_hub library:

from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id="Miguel764/efficientnetv2s-skin-cancer-classifier",
    filename="efficientnetv2s.h5"
)

---

License: MIT

---