Skin Lesion Classification Using CNN

Overview

This project presents a deep learning-based skin lesion classification framework using a Custom Convolutional Neural Network (CNN) implemented in PyTorch. The model is trained on the HAM10000 dataset to classify dermoscopic images into seven different skin lesion categories.

The project also includes:

• Stratified dataset splitting (70/15/15)
• Training and validation pipeline
• Performance evaluation metrics
• Confusion matrix generation
• Accuracy and loss visualization curves
• Trained model saving

---

Dataset

Dataset Used: HAM10000 Skin Lesion Dataset

Dataset Link: https://www.kaggle.com/datasets/kmader/skin-cancer-mnist-ham10000

Classes:

• MEL (Melanoma)
• NV (Melanocytic Nevi)
• BCC (Basal Cell Carcinoma)
• AKIEC (Actinic Keratosis)
• BKL (Benign Keratosis)
• DF (Dermatofibroma)
• VASC (Vascular Lesions)

---

Technologies Used

• Python
• PyTorch
• NumPy
• Pandas
• Matplotlib
• Seaborn
• Scikit-learn

---

Model Details

• Custom CNN Architecture
• Optimizer: Adam
• Loss Function: CrossEntropyLoss
• Epochs: 5
• Batch Size: 32
• Input Size: 64x64

---

Performance Metrics

Metric	Value
Accuracy	65.00%
Precision	76.59%
Recall	65.00%
F1 Score	68.49%
AUC-ROC	78.83%

---

Project Structure

SkinLesion-MobileNetV4/
│
├── dataset/
├── results/
│   ├── accuracy_curve.png
│   ├── loss_curve.png
│   └── confusion_matrix.png
│
├── best_model.pth
├── requirements.txt
├── train.py
└── README.md

---

How to Run

Install Dependencies

pip install -r requirements.txt

Run Training

python train.py

---

Output

The project generates:

• Trained model weights (best_model.pth)
• Accuracy curve
• Loss curve
• Confusion matrix
• Final evaluation metrics

---

Authors

• Nitesh Kumar (23BCS058)
• Pranav Praveen (23BCS067)
• Gaurav Kumar (23BCS031)

Department of Computer Science and Engineering Shri Mata Vaishno Devi University (SMVDU) Katra, Jammu & Kashmir, India