文件列表:

models/
outputs/
scripts/

# Dermoscopy Image Classification for Melanoma Detection This project implements a deep learning approach for classifying dermoscopy images, aiding in the diagnosis of melanoma. The project builds upon the master thesis titled ["Combining deep learning, handcrafted features, and metadata for the classification of dermoscopy images"](https://dione.lib.unipi.gr/xmlui/handle/unipi/15875). The proposed method involves the use of a combination of EfficientNet-B0 features, GLCM, LBP, color moments features, and metadata features to improve base model's performance by **93,97% AUC score**. ## Project Structure The project is organized into the following directories: - **models**: This directory stores the trained model weights generated during training. - **outputs**: This directory stores any output files generated during training or validation, such as loss curves, visualizations, or predictions. - **scripts**: This directory contains Python modules (.py files) that are imported and used by other scripts for various functionalities: - Training and validation (`train-val.ipynb`) - Data handling (`dataio.py`, `prepareDataset.ipynb`) - Preprocessing (`preprocess.py`) - Feature extraction (`colorfeatures.py`, `feature_extraction.py`) - Model prediction (`classify.py`, `predict.ipynb`) - Data augmentation (`dataAugmentation.ipynb`) - Metadata analysis (`metadataAnalysis.ipynb`) - Segmentation (`segment.py`) - Testing (`test.ipynb`) ## Methodology ![screenshot](https://github.com/DimitrisTsel/Dermoscopy-Image-Classification/blob/main/outputs/image.png?raw=true") ## Dependencies Python libraries: - TensorFlow - Keras - Scikit-learn - OpenCV - NumPy - matplotlib - seaborn ## Data The project uses a dermoscopy image dataset of [The ISIC 2020 Challenge](https://challenge2020.isic-archive.com/) for Skin Lesion Analysis Towards Melanoma Detection. ## Functionality ### Data Preparation: - Load the dermoscopy image dataset (imports `dataio.py`). - Preprocess the images (e.g., resizing, normalization) (imports `preprocess.py`). - Split the data into training, validation, and testing sets (refer to `prepareDataset.ipynb`). - Implement data augmentation techniques to balance Benign and Malignant classes and help a model improve the accuracy of predictions (refer to `dataAugmentation.ipynb`). ### Feature Extraction: - Extract color features from images (potentially using `colorfeatures.py`). - Extract handcrafted features, such as Grey Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP) features (imports `feature_extraction.py`). - Pre-trained deep learning models like DenseNet, VGG19, and EfficientNet for feature extraction. - Combine these features (imports `feature_extraction.py`). ### Model Training: - Define a convolutional neural network (CNN) architecture suitable for image classification (refer to `train-val.ipynb`). - Integrate the extracted features, including deep learning features from DenseNet, VGG19, and EfficientNet, handcrafted features like GLCM and LBP, and patients' metadata features, into the model. - Train the model on the training set, monitoring progress using metrics like accuracy and loss. - Implement techniques like early stopping to prevent overfitting. ### Validation and Testing: - Evaluate the trained model's performance on the validation set (refer to `train-val.ipynb`). - Optionally, evaluate the model on a separate test set (refer to `test.ipynb`) for a more unbiased assessment. ### Evaluation: Calculates and display the following metrics: - Accuracy: Proportion of correctly classified samples. - Precision: Ratio of true positives to all predicted positives. - Recall: Ratio of true positives to all actual positives. - F1 Score: Harmonic mean of precision and recall, useful when dealing with imbalanced class distributions. - Specificity: Ratio of true negatives to all predicted negatives. - AUC (Area Under the ROC Curve): Represents the probability that the model ranks a randomly chosen positive instance higher than a randomly chosen negative instance. The notebook provides visualizations of these metrics, such as confusion matrices or ROC curves, to aid in model evaluation. ### Prediction: - Use the trained model to classify new dermoscopy images. - The model outputs the predicted class (e.g., melanoma or benign). ## Using Jupyter Notebooks for Dermoscopy Image Classification ### Option 1: Using a Jupyter Notebook server 1. **Install a Jupyter Notebook server** following the instructions for your operating system. 2. **Navigate to the project directory** in your terminal: ```bash cd Dermoscopy-Image-Classification ``` 3. **Start the Jupyter Notebook server**: ```bash jupyter notebook ``` This will open a web interface in your browser, typically at [http://localhost:8888](http://localhost:8888). 4. From the list of notebooks, open the script`train-val.ipynb`. ### Option 2: Using a local kernel (command line) 1. Ensure you have a Python kernel installed for Jupyter Notebooks (e.g., ipykernel). 2. **Activate the project's virtual environment** (if applicable). 3. **Navigate to the project directory** in your terminal. 4. **Run the desired script directly**: ```bash jupyter notebook train-val.ipynb ``` ## Further Development Here are some suggestions for further development: - Experiment with different CNN architectures and more powerful models with more parameters. - Explore advanced training techniques like transfer learning. - Explore explainability in Deep Learning and analyze the decisions behind the model's predictions.

近期下载者：

相关文件：

评论：[我要评论] [举报此文件]

收藏者：