MoleMonitoring: Skin Mole Classification

MoleMonitoring is a deep learning project from the Samsung Innovation Campus, focused on classifying skin moles using dermoscopic images. Our best model, Fusion Model + SVM, achieved 91% accuracy on HAM10000 and 83% accuracy on ISIC2018, outperforming baseline models like ResNet-50.

MoleMonitoring Pipeline Overview

1. Beginning	2. Empty	3. Gallery
4. Analyze Mole	5. View Results	6. Example Malignant

Team Members:

• Manuel Pinto - GitHub: ManuelCPinto
• André Branco - GitHub: Aser28860d
• Francisco Silva - GitHub: fpgsilva
• João Pedro Silveira - GitHub: Joao-Pedro-Silveira

The HAM10000 dataset includes the following seven classes of skin lesions, categorized into Benign and Malignant:

Malignant Lesions

• Actinic Keratoses (akiec)
  Precancerous skin lesions that may develop into squamous cell carcinoma.

• Basal Cell Carcinoma (bcc)
  A common and slow-growing type of skin cancer that rarely spreads.

• Melanoma (mel)
  An aggressive and potentially fatal form of skin cancer that develops in melanocytes.

Benign Lesions

• Benign Keratosis-like Lesions (bkl)
  Non-cancerous growths, including seborrheic keratoses and solar lentigines.

• Dermatofibroma (df)
  Small, firm nodules that are benign skin tumors.

• Melanocytic Nevi (nv)
  Commonly known as moles or beauty marks, these are benign accumulations of melanocytes.

• Vascular Lesions (vasc)
  Benign abnormalities of blood vessels, such as hemangiomas or angiomas.

Dataset Preparation

Follow these steps to download and prepare the HAM10000 dataset for this project:

1. Download the Dataset

1. Visit the HAM10000 dataset page.
2. Download the file named dataverse_files.zip.
3. Place the downloaded file in the root directory of the project.

2. Run the Preprocessing Scripts

1. Run the following commands on the project directory:

   python zipMerge.py
   python update_metadata.py
   python process_images.py
   python generate_plots.py

3. Output Structure

MoleMonitoring/
├── HAM10000/
│   ├── HAM10000_images/               # Resized images (256x256, not normalized)
│   ├── HAM10000_images_processed/     # Processed images (normalized)
│   │   ├── rgb/                       # Resized + normalized RGB images (256x256, [0, 1])
│   │   ├── grayscale/                 # Resized + normalized grayscale images (256x256, [0, 1])
│   ├── HAM10000_metadata/
├── ISIC2018/
│   ├── ISIC2018_images/               # Resized images (256x256, not normalized)
│   ├── ISIC2018_images_processed/     # Processed images (normalized)
│   │   ├── rgb/                       # Resized + normalized RGB images (256x256, [0, 1])
│   │   ├── grayscale/                 # Resized + normalized grayscale images (256x256, [0, 1])
│   ├── ISIC2018_metadata/
├── models/
│   ├── baseline.py
│   ├── cnn_with_weights.py
│   ├── efficientnet-resnet-vit-svm.ipynb
│   ├── irv2-sa.ipynb
│   ├── resnet50.py
├── molemonitoringapp/
├── pictures/
│   ├── gallery.png
│   ├── nvMole.png
│   ├── homepage.png
├── plots/
├── preprocessing/
│   ├── fitzpatrick.py
│   ├── generate_plots.py
│   ├── process_images.py
│   ├── update_metadata.py
│   ├── zip_merge.py
├── vertex/
│   ├── app.py
│   ├── Dockerfile
│   ├── requirements.txt
├── .gitignore
├── README.md