Browsing by Author "Asad Ullah, Midhat Basit and Umer Ashiq"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Skin acne detection system using deep learning algorithms
    (UMT, Lahore, 2023) Asad Ullah, Midhat Basit and Umer Ashiq
    Acne being one of the very common skin conditions affect a huge population worldwide. Early diagnosis and treatment of acne can help lessen or even eradicate the chances of long-term scarring. In recent years, the implication of deep learning algorithms in medical image analysis has shown promising results. In this paper, we have proposed a skin acne detection system using Deep learning algorithm. This paper presents the development of a system utilizing Convolutional Neural Networks (CNNs) followed by validation of the results produced by another CNN architecture. The system's architecture was trained on a dataset encompassing various types of acne, including blackhead, pustules, papules, milia, and cysts. The dataset of 6548 was augmented to a total of 11,483 images to improve diversity and generalizability. The CNN model achieved remarkable results, with a training loss 12.3%, training accuracy 95.5%, validation loss 10.2%, validation accuracy 97%, f1 score 0.973 and recall 0.972. Moreover, to further enhance the system's performance, the VGG 16 and VGG-19 architecture was trained on the same dataset, yielding a training loss 45.3%, training accuracy 82.4%, validation loss 37%, validation accuracy 86.9%, f1 score 0.862 and recall 0.86 along with training loss 54.8%, training accuracy 79.3 %, validation loss 44.75%, validation accuracy 84.2%, f1 score of 0.832 and recall 0.83. These results demonstrate the effectiveness of the proposed system in accurately detecting and classifying various types of acne lesions. The outcomes also highlight the importance of dataset diversity and the utilization of different deep learning algorithms for improved performance.