Browsing by Author "Islam, Sayma"

Now showing 1 - 2 of 2

A Shallow Deep Learning Approach To Classify Skin Cancer Using Down-Scaling Method To Minimize Time and Space Complexity
(Daffodil International University, 2022-08-04) Sidratul, Montaha; Azam, Sami; Rafid, A. K. M. Rakibul Haque; Islam, Sayma; Ghosh, Pronab; Jonkman, Mirjam
The complex feature characteristics and low contrast of cancer lesions, a high degree of inter-class resemblance between malignant and benign lesions, and the presence of various artifacts including hairs make automated melanoma recognition in dermoscopy images quite challenging. To date, various computer-aided solutions have been proposed to identify and classify skin cancer. In this paper, a deep learning model with a shallow architecture is proposed to classify the lesions into benign and malignant. To achieve effective training while limiting over fitting problems due to limited training data, image preprocessing and data augmentation processes are introduced. After this, the ‘box blur’ down-scaling method is employed, which adds efficiency to our study by reducing the overall training time and space complexity significantly. Our proposed shallow convolutional neural network (SCNN_12) model is trained and evaluated on the Kaggle skin cancer data ISIC archive which was augmented to 16485 images by implementing different augmentation techniques. The model was able to achieve an accuracy of 98.87% with optimizer Adam and a learning rate of 0.001. In this regard, parameter and hyper-parameters of the model are determined by performing ablation studies. To assert no occurrence of overfitting, experiments are carried out exploring k-fold cross-validation and different dataset split ratios. Furthermore, to affirm the robustness the model is evaluated on noisy data to examine the performance when the image quality gets corrupted. This research corroborates that effective training for medical image analysis, addressing training time and space complexity, is possible even with a light weighted network using a limited amount of training data.
An Analysis of Depp Learning Approaches with Image Preprocessing Technique To Predict Monkeypox
(Daffodil International University, 23-02-12) Biswas, Ritu; Islam, Sayma
As the world continues to recover from COVID-19, the virus monkeypox provides a new pandemic intimidation. Monkeypox isn't as deadly or as widespread as the virus of COVID-19, but some new cases are reported regularly from the several nations. Without sufficient precautions, it shouldn't come as a fact if another worldwide pandemic occurs. The efficiency of an automated deep-learning algorithm for categorization was evaluated using the monkeypox dataset. The main objective is to create a deep learning model that will understand the different datasets as accurately as appropriate. First, similar image preprocessing techniques are applied to enhance brightness and contrast for this dataset. As the datasets utilized in this experiment contain few images to effectively train a deep learning model, data augmentation is applied. Then, a VGG-16 model with randomly selected hyperparameters and layer numbers is generated. VGG16, VGG19, MobileNetV2, and ResNet50 are evaluated to determine whether the model yielded the highest performance. VGG16 is utilized as the foundational model because it has the best accuracy. In this work, VGG-16 was utilized to assess its robustness and get the highest accuracy level obtainable. Using image preprocessing algorithms with appropriate parameter values, image quality is enhanced. A total of 770 preprocessed images of monkeypox were augmented using seven approaches, yielding a collection of 1991 images. The model was then evaluated for its robustness. Results were compared to those obtained from previous research. Within a training accuracy of 97.29%, a validation accuracy of 95.26%, and a test accuracy of 96.29%, the VGG-16 model performed the best. VGG19 achieved test accuracy of 95.06%, MobileNetV2 of 94.59%, and ResNet50 of 54.32%. Our proposed algorithm, which relies on preprocessing of images, transfer learning, and adjustment, which shown a high level of accuracy in identifying monkeypox based on a small set of intricate photos. After determining appropriate configuration, this model is trained by using remaining of the data set to evaluate overall performance. We generate and evaluate different performance measures for the monkeypox dataset, including accuracy, precision, recall, specificity, and F1-score.