Conference Papers
Permanent URI for this collectionhttps://idr.nitk.ac.in/handle/123456789/28506
Browse
Search Results
Item Diabetic Retinopathy Severity Classification based on attention mechanism(Institute of Electrical and Electronics Engineers Inc., 2023) Jha, A.; Ananthanarayana, V.S.One of the significant factors causing blindness is diabetic retinopathy, a typical microvascular side effect of diabetes. Highly qualified professionals often examine colored fundus photos to identify this catastrophic condition. It takes much time and effort for ophthalmologists to diagnose diabetic retinopathy (DR) manually. The number of diabetes patients has dramatically increased during the last several years, which has made automated DR diagnosis a research hotspot. This paper proposes a hybrid deep learning model using a pre-trained DenseNet architecture integrated with CBAM for feature refinement. The dataset provided by the Kaggle Asia Pacific Tele-Ophthalmology Society (APTOS), having 3662 fundus images, is used in this research. In the multiclass classification experiment, we achieved 86.22% accuracy and 91.44 Kappa score(QWK). The local interpretable model-agnostic explanations (LIME) framework is used to assess predictions further and produce visual explanations, which can assist in decreasing the drawback of black-box models in aiding medical decision-making. © 2023 IEEE.Item An Interpretable Deep Learning Model for Skin Lesion Classification(Springer Science and Business Media Deutschland GmbH, 2023) Jha, A.; Ananthanarayana, V.S.Skin Cancer is a dangerous issue in society. Early diagnosis and therapy are two of the most crucial steps in preventing the onset of a disease. Dermatologists primarily use visual methods to identify the skin lesions which may cause skin cancer. With the development of technology, methods for classifying skin lesions, like deep learning and computer vision, are gaining popularity. A hybrid model is proposed using a pre-trained DenseNet architecture integrated with Convolutional Block Attention Module (CBAM) for feature refinement. The HAM10000 dataset, which includes 10015 dermoscopic pictures with seven distinct skin disease types, was used in our research. The proposed approach outperforms the original pre-trained DenseNet models, with an average accuracy of 93%. The LIME framework is used to assess predictions further and produce visual explanations, which can assist in decreasing the drawback of black-box models in aiding medical decision-making. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.Item Data Augmentation vs. Synthetic Data Generation: An Empirical Evaluation for Enhancing Radiology Image Classification(Institute of Electrical and Electronics Engineers Inc., 2023) Shetty, S.; Ananthanarayana, V.S.; Mahale, A.Radiology is a field of medicine dealing with diagnostic images to detect diseases for further treatment. Collecting and annotating diagnostic images like Magnetic Resonance Imaging (MRI) and X-Ray is a rigorous and time-consuming process. Deep Learning methods are widely utilized for disease classification and prediction from diagnostic images, but they demand substantial amounts of training data. Additionally, certain diseases are uncommon in large patient cohorts, posing difficulties in obtaining sufficient imaging samples to construct accurate deep learning models. Data augmentation techniques are commonly used to overcome this challenge of limited data. These techniques involve applying geometric transformations such as rotation, cropping, zooming, flipping, and other similar operations to the images to enlarge the dataset artificially. Another possible way of expanding the dataset is by synthesizing data to generate artificial medical images by mimicking the original images. This study presents RAD-DCGAN: A Deep Convolutional Generative Adversarial Network to produce high-resolution synthetic radiology images from the X-ray and MRI images collected from a private medical hospital (KMC Hospital, India). We aim to determine the most effective technique for enhancing the performance of radiology image classifiers by comparing and evaluating the proposed RAD-DCGAN with the standard data augmentation strategy. Our empirical evaluation, which involved eight standard deep learning models, demonstrated that deep learning classifiers trained on synthetic radiology data outperformed those trained on standard augmented data. The utilization of the RAD-DCGAN model for training and testing deep learning models on synthetic data has demonstrated a notable improvement of 4-5% in accuracy compared to conventional augmentation techniques. This signifies the state-of-the-art performance achieved by the RAD-DCGAN model in enhancing the accuracy of deep learning models. © 2023 IEEE.