Skin Cancer Classification with DenseNet Deep Convolutional Neural Network

Abstract

Skin cancer has been a growing health concern with a significant impact on survival rates. Accurate diagnosis plays a crucial role in effective treatment. To address this, a DenseNet model is proposed for classifying different types of skin cancer using the HAM10000 Dataset, which consists of 10,015 skin lesion photos categorized into six types. To address the class imbalance, the dataset is pre-processed and oversampling techniques are applied. The images are resized to a standardized 32x32 pixel format to improve model performance. The model builds upon the pre-trained DenseNet121 architecture originally trained on the ImageNet dataset. Additional fully connected layers are incorporated to create a sequential model. Training is performed using the Adam optimizer and a sparse categorical cross-entropy loss function. These techniques are essential for optimizing model performance and ensuring accurate classification of skin lesions. Evaluation of the model’s performance is based on the accuracy metric, and it has demonstrated high accuracy on the HAM10000 Dataset, effectively capturing important features of skin lesions. The proposed model holds great potential for dermatologists in detecting skin cancer and planning appropriate treatment strategies for patients with its high accuracy and reliable classification capabilities.