Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
8 results
Search Results
Item Infant Brain MRI Segmentation Using Deep Volumetric U-Net with Gamma Transformation(Springer Science and Business Media Deutschland GmbH, 2023) Yeshwanth, G.S.; Annappa, B.; Dodia, S.; Manoj Kumar, M.V.The growth of the brain from infantile to adolescence is very complex and takes a very long period. There are many processes such as myelination, migration, neural induction, and many other time-taking processes to study the development of the brain. This makes it necessary to develop some automatic tools to study the development of the brain. The brain consists mainly of three parts white matter, gray matter, and cerebrospinal fluid. So, quantitative tools will be a great boon for the medical community to deal with the brain if the brain MRI images are segmented into these three different parts. Although there are some tools for segmenting adult MRI images, for 6-month child segmentation, the brain becomes challenging as the white matter and gray matter are almost indistinguishable due to the brain development process. Segmentation of brain MRI images can identify specific patterns that contribute to healthy brain development. The dataset to address this problem had been taken from the Iseg2019 challenge conducted by MICCAI. Segmentation of MRI needs expert doctors. Advancements in computer vision techniques can be used to replace present time-consuming work. This paper proposes a deep learning model for image segmentation using a three-dimensional U-net. The proposed model gives dice values of 93.75, 88.24, and 85.64 for cerebrospinal fluid, gray matter, and white matter. This paper also presents various experimental results of U-net, attention U-net with different modifications. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item Automatic Abnormality Detection in Musculoskeletal Radiographs Using Ensemble of Pre-trained Networks(Springer Science and Business Media Deutschland GmbH, 2023) Verma, R.; Jain, S.; Saritha, S.K.; Dodia, S.Musculoskeletal disability (MSDs) defined as the injuries that affect the movement or musculoskeletal system of the human body. Over the worldwide, it is the second most cause of physical disability. Musculoskeletal disability worsens over time and can result in long-term discomfort and severe disability. As a result, early detection and diagnosis of these anomalies is essential. But the diagnosis process is very time consuming, error prone and required diagnostic professional. Deep learning algorithms have recently been applied in medical imaging that provides a robust platform with very reliable outcomes. The development of Computer Aided Detection (CAD) system extensively speed up the diagnosis process. In this paper, a weighted ensemble model has been proposed, which is the combination of three pre-trained models (DenseNet169, MobileNet, and XceptionNet). The weighted ensemble model is tested on MURA dataset, a large public dataset provided by Stanford ML Group. Our model achieved a cohen’s kappa score 0.739 with precision of 0.885 and recall of 0.854, which is higher than many existing approaches such as densenet169 and ensemble200 model. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item Classification of Skin Cancer Images using Lightweight Convolutional Neural Network(Institute of Electrical and Electronics Engineers Inc., 2023) Sandeep Kumar, T.; Annappa, B.; Dodia, S.Skin is the most powerful shield human organ that protects the internal organs of the human body from external attacks. This important organ is attacked by a diverse range of microbes such as viruses, fungi, and bacteria causing a lot of damage to the skin. Apart from these microbes, even dust plays important role in damaging skin. Every year several people in the world are suffering from skin diseases. These skin diseases are contagious and spread very fast. There are varieties of skin diseases. Thus it requires a lot of practice to distinguish the skin disease by the doctor and provide treatment. In order to automate this process several deep learning models are used in recent past years. This paper demonstrates an efficient and lightweight modified SqueezeNet deep learning model on the HAM10000 dataset for skin cancer classification. This model has outperformed state-of-the-art models with fewer parameters. As compared to existing deep learning models, this SqueezeNet variant has achieved 99.7%, 97.7%, and 97.04% as train, validation, and test accuracies respectively using only 0.13 million parameters. © 2023 IEEE.Item An Efficient Deep Transfer Learning Approach for Classification of Skin Cancer Images(Springer Science and Business Media Deutschland GmbH, 2023) Naik, P.P.; Annappa, B.; Dodia, S.Prolonged exposure to the sun for an extended period can likely cause skin cancer, which is an abnormal proliferation of skin cells. The early detection of this illness necessitates the classification of der-matoscopic images, making it an enticing study problem. Deep learning is playing a crucial role in efficient dermoscopic analysis. Modified version of MobileNetV2 is proposed for the classification of skin cancer images in seven classes. The proposed deep learning model employs transfer learning and various data augmentation techniques to more accurately classify skin lesions compared to existing models. To improve the per¬formance of the classifier, data augmentation techniques are performed on “HAM10000" (Human Against Machine) dataset to classify seven dif¬ferent kinds of skin cancer. The proposed model obtained a training accuracy of 96.56% and testing accuracy of 93.11%. Also, it has a lower number of parameters in comparison to existing methods. The aim of the study is to aid dermatologists in the clinic to make more accurate diagnoses of skin lesions and in the early detection of skin cancer. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.Item Optimizing Super-Resolution Generative Adversarial Networks(Springer Science and Business Media Deutschland GmbH, 2023) Jain, V.; Annappa, B.; Dodia, S.Image super-resolution is an ill-posed problem because many possible high-resolution solutions exist for a single low resolution (LR) image. There are traditional methods to solve this problem, they are fast and straightforward, but they fail when the scale factor is high or there is noise in the data. With the development of machine learning algorithms, their application in this field is studied, and they perform better than traditional methods. Many Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) have been developed for this problem. The Super-Resolution Generative Adversarial Networks (SRGAN) have proved to be significant in this area. Although the SRGAN produces good results with 4 upscaling, it has some shortcomings. This paper proposes an improved version of SRGAN with reduced computational complexity and training time. The proposed model achieved an PPSNR of 29.72 and SSIM value of 0.86. The proposed work outperforms most of the recently developed systems. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item COVID-19: Automatic detection from X-ray images by utilizing deep learning methods(Elsevier Ltd, 2021) Nigam, B.; Nigam, A.; Jain, R.; Dodia, S.; Arora, N.; Annappa, B.In recent months, a novel virus named Coronavirus has emerged to become a pandemic. The virus is spreading not only humans, but it is also affecting animals. First ever case of Coronavirus was registered in city of Wuhan, Hubei province of China on 31st of December in 2019. Coronavirus infected patients display very similar symptoms like pneumonia, and it attacks the respiratory organs of the body, causing difficulty in breathing. The disease is diagnosed using a Real-Time Reverse Transcriptase Polymerase Chain reaction (RT-PCR) kit and requires time in the laboratory to confirm the presence of the virus. Due to insufficient availability of the kits, the suspected patients cannot be treated in time, which in turn increases the chance of spreading the disease. To overcome this solution, radiologists observed the changes appearing in the radiological images such as X-ray and CT scans. Using deep learning algorithms, the suspected patients’ X-ray or Computed Tomography (CT) scan can differentiate between the healthy person and the patient affected by Coronavirus. In this paper, popular deep learning architectures are used to develop a Coronavirus diagnostic systems. The architectures used in this paper are VGG16, DenseNet121, Xception, NASNet, and EfficientNet. Multiclass classification is performed in this paper. The classes considered are COVID-19 positive patients, normal patients, and other class. In other class, chest X-ray images of pneumonia, influenza, and other illnesses related to the chest region are included. The accuracies obtained for VGG16, DenseNet121, Xception, NASNet, and EfficientNet are 79.01%, 89.96%, 88.03%, 85.03% and 93.48% respectively. The need for deep learning with radiologic images is necessary for this critical condition as this will provide a second opinion to the radiologists fast and accurately. These deep learning Coronavirus detection systems can also be useful in the regions where expert physicians and well-equipped clinics are not easily accessible. © 2021 Elsevier LtdItem A novel receptive field-regularized V-net and nodule classification network for lung nodule detection(John Wiley and Sons Inc, 2022) Dodia, S.; Annappa, B.; Mahesh, M.Recent advancements in deep learning have achieved great success in building a reliable computer-aided diagnosis (CAD) system. In this work, a novel deep-learning architecture, named receptive field regularized V-net (RFR V-Net), is proposed for detecting lung cancer nodules with reduced false positives (FP). The method uses a receptive regularization on the encoder block's convolution and deconvolution layer of the decoder block in the V-Net model. Further, nodule classification is performed using a new combination of SqueezeNet and ResNet, named nodule classification network (NCNet). Postprocessing image enhancement is performed on the 2D slice by increasing the image's intensity by adding pseudo-color or fluorescence contrast. The proposed RFR V-Net resulted in dice similarity coefficient of 95.01% and intersection over union of 0.83, respectively. The proposed NCNet achieved the sensitivity of 98.38% and FPs/Scan of 2.3 for 3D representations. The proposed NCNet resulted in considerable improvements over existing CAD systems. © 2021 Wiley Periodicals LLC.Item KAC SegNet: A Novel Kernel-Based Active Contour Method for Lung Nodule Segmentation and Classification Using Dense AlexNet Framework(World Scientific, 2024) Dodia, S.; Annappa, B.; Mahesh, P.A.Lung cancer is known to be one of the leading causes of death worldwide. There is a chance of increasing the survival rate of the patients if detected at an early stage. Computed Tomography (CT) scans are prominently used to detect and classify lung cancer nodules/tumors in the thoracic region. There is a need to develop an efficient and reliable computer-aided diagnosis model to detect lung cancer nodules accurately from CT scans. This work proposes a novel kernel-based active-contour (KAC) SegNet deep learning model to perform lung cancer nodule detection from CT scans. The active contour uses a snake method to detect internal and external boundaries of the curves, which is used to extract the Region Of Interest (ROI) from the CT scan. From the extracted ROI, the nodules are further classified into benign and malignant using a Dense AlexNet deep learning model. The key contributions of this work are the fusion of an edge detection method with a deep learning segmentation method which provides enhanced lung nodule segmentation performance, and an ensemble of state-of-the-art deep learning classifiers, which encashes the advantages of both DenseNet and AlexNet to learn better discriminative information from the detected lung nodules. The experimental outcome shows that the proposed segmentation approach achieves a Dice Score Coefficient of 97.8% and an Intersection-over-Union of 92.96%. The classification performance resulted in an accuracy of 95.65%, a False Positive Rate, and False Negative Rate values of 0.0572 and 0.0289. The proposed model is robust compared to the existing state-of-the-art methods. © 2024 World Scientific Publishing Company.