[Project Addition]: Skin Cancer Detection using DL

[Sindhu-2004]

A skin cancer detection model using TensorFlow to classify images as malignant or benign.

[abhisheks008]

Can you share the dataset and approach for solving this issue?

[Sindhu-2004]

Project Title : Skin Cancer Detection Aim : A skin cancer detection model using TensorFlow to classify images as malignant or benign. Dataset : [Skin Cancer ISIC] https://www.kaggle.com/datasets/nodoubttome/skin-cancer9-classesisic Approach : This model is implemented using TensorFlow, Convolutional Neural Network, Seaborn, Matplotlib, PIL, scikit-learn

[abhisheks008]

Hi @Sindhu-2004 can you please share your deep learning approach a bit briefly? Need some clarification in the architectures/methods/models you are trying to implement here.

[jeet-Abhi123]

Hi! @abhisheks008 I would like to do this project. Dataset : HAM10000 (7 classes of skin lesions) Approach : For image preprocessing I will be experimenting with dull razor, k-means or YOLOv4. After that I will be applying Transfer learning techniques like Resnet or Mobilenet and adding my own customized ANN layers after that. Finally I will apply KerasTuner for Hyperparameters and plot the graphs for results.

[abhisheks008]

Hi! @abhisheks008 I would like to do this project. Dataset : HAM10000 (7 classes of skin lesions) Approach : For image preprocessing I will be experimenting with dull razor, k-means or YOLOv4. After that I will be applying Transfer learning techniques like Resnet or Mobilenet and adding my own customized ANN layers after that. Finally I will apply KerasTuner for Hyperparameters and plot the graphs for results.

But this issue is raised by another contributor, you can look for the issues which are raised by me.

[jeet-Abhi123]

@abhisheks008 But your issues has been assigned to someone else.

[abhisheks008]

@abhisheks008 But your issues has been assigned to someone else.

Not all the issues have been assigned. Check out in the issues section, many of them are not assigned.

[Sindhu-2004]

Approach : In this skin cancer detection project we implement a convolutional neural network (CNN) with the EfficientNetB7 architecture, leveraging transfer learning for improved performance. By loading a dataset of skin images labeled as malignant or benign, using Pandas and Numpy for data handling and Matplotlib for visualization. The dataset is split into training and validation sets, with images preprocessed by resizing and normalizing pixel values. EfficientNetB7, pre-trained on ImageNet, serves as the backbone for feature extraction, with its layers frozen to retain learned weights. The model is built using Keras Functional API, incorporating layers for flattening, dense connections with 256 units each, ReLU activation functions for non-linearity, Batch Normalization layers for training stabilization, and a Dropout layer with a 0.3 rate to prevent overfitting. The final layer is a single neuron with a sigmoid activation function, producing a probability score for malignancy. The model is compiled using the Adam optimizer and binary cross-entropy loss, with AUC as the evaluation metric. Training is conducted over multiple epochs, followed by performance evaluation through plots of training and validation loss and AUC metrics, demonstrating the model's effectiveness in distinguishing between malignant and benign skin lesions. This approach showcases the utility of transfer learning and advanced CNN architectures in medical image classification tasks.

[abhisheks008]

Along side EfficientNet can you implement 2 more architectures for this project?

[Sindhu-2004]

I would like to add architectures like VGG-16 to develop a model, which contributes to the improvement of accuracy and ResNet (Residual Neural Network) which is used to obtain the diagnosis report as a confidence score with high accuracy.

[abhisheks008]

Cool then VGG, EfficientNet and ResNet are the final models right? @Sindhu-2004

[Sindhu-2004]

Yes sir

[abhisheks008]

Assigned @Sindhu-2004