@nits.ac.in
ASSISTANT PROFESSOR GRADE I
NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR
IMAGE AND VIDEO PROCESSING
MEDICAL IMAGE PROCESSING
DATA SCIENCE
MACHINE LEARNING
Scopus Publications
Scholar Citations
Scholar h-index
Scholar i10-index
S. Berlin Shaheema, Suganya Devi K., and Naresh Babu Muppalaneni
Elsevier BV
Prasenjit Dhar, K. Suganya Devi, Satish Kumar Satti, and P. Srinivasan
Springer Science and Business Media LLC
Hemanth Kumar Vasireddi, Suganya Devi K, and G. N. V. Raja Reddy
Springer Science and Business Media LLC
K. Sekar, K. Suganya Devi, T. Dheepa, and P. Srinivasan
Wiley
Sekar K., Suganya Devi K., Satish Kumar Satti, and Srinivasan P.
Elsevier BV
Satish Kumar Satti, Suganya Devi K, and Srinivasan P
Wiley
Himanshi Singh, K Suganya Devi, Shraddha Singh Gaur, and Ramanuj Bhattacharjee
IEEE
Skin cancer is a prevalent and potentially fatal disease that affects a large number of individuals worldwide. Detecting skin cancer early is vital for effective treatment and positive patient outcomes. In this research, a deep learning (DL) model is presented that utilizes the VGG19 architecture with self-attention modules to automate skin cancer detection. The proposed model achieved an impressive overall accuracy of 91.4% on HAM10000 dataset and outperformed several standard classification models, proving that incorporating self-attention modules into convolutional neural networks can significantly enhance skin cancer detection systems' efficiency. Nonetheless, the model has certain limitations, such as a limited dataset and lower accuracy for the melanoma class. Thus, further studies are required to improve the model's performance, which could potentially save lives by allowing for early detection and treatment of melanoma. This proposed approach has shown promising results for automating skin cancer detection, and the potential impact of this research is significant for developing more accurate and reliable skin cancer detection systems.
Ramanuj Bhattacharjee, K Suganya Devi, and S. Vijaykanth
IEEE
To improve the chances of survival for a patient with laryngeal cancer, early detection is crucial. Currently, the standard diagnostic method involves an endoscopic examination of the larynx, followed by a biopsy and histological analysis by an oncologist, which can be subject to variability due to subjective evaluation. Therefore, there is a need for a faster and more accurate detection system that can replace the current manual examination. Recent research has shown that Deep Learning technology can assist in identifying laryngeal cancer, including precancerous and cancerous tumors, from endoscopic pictures. However, endoscopic image processing is a challenging task due to the highly dynamic nature of the endoscopic video, spectrum fluctuations, and numerous image interferences. To address this challenge, a Deep Ensemble Learning approach using convolutional neural networks (CNNs) and an effective image segmentation technique has been proposed. The suggested model has an overall accuracy of 98.12%.
Prasenjit Dhar, K Suganya Devi, K Sekar, and P Srinivasan
IEEE
Poikilocytosis is a condition in which the shape of RBC cells changes regularly. This condition is arising from several anemias and other diseases. It is critical to detect and classify abnormal RBC shapes as early as possible to diagnose and treat the condition. The proposed four different deep Convolutional Neural Networks (CNNs) are for RBC morphology classification, which is beneficial to hematologists. After using the data augmentation strategy, deep CNN produces good results and improves the performance of the models. The proposed deep CNN is tested on the recently available Erythrocytes IDB datasets. In Erythrocytes, there are three versions of IDB datasets IDB 1, IDB 2, and IDB 3. The proposed deep CNN models obtained an accuracy of 97.77 %, 96.28 %, 96.10 %, and 95.95 % and outperformed existing benchmark methods. The proposed CNN models produce good accuracy, speed, and classification rate results.
Prasenjit Dhar, K Suganya Devi, and P Srinivasan
Informa UK Limited
Sekar K, Suganya Devi K, J. Suganthi, and Dheepa T
IEEE
Developing energy-efficient data collection protocols is a big challenge for wireless sensor networks (WSN) since sensor nodes are battery-powered and have severe energy limitations. Clustering methods minimize the power consumption of every sensor. Hot spots, however, are an issue in areas near the sink. Modern developments indicate the substantial development of nature-inspired optimization-based routing strategies used for data collection in WSNs to minimize energy consumption and extend lifespan. This paper presents a novel Jellyfish Search Algorithm (JSA) based on an optimal routing protocol for data aggregation in WSN. Firstly, the proposed method optimizes the size of the cluster, and secondly, a routing protocol is introduced for efficient data transmission. The simulation results show that the proposed method achieves better energy-efficient data collection and increases the lifespan of sensors.
NV Rajareddy Goluguri, Suganya Devi K, and Prathima CH
ENGG Journals Publications
Satish Kumar Satti, Suganya Devi K, and Srinivasan P
Wiley
Ramanuj Bhattacharjee, Suganya Devi K, Sekar K, and Dheepa T
IEEE
The main cause of mortality globally is cancer, with lung cancer being the most common among all cancer types. Radiologists use computer tomography (CT) scans to detect and monitor cancer in the body. Due to subjectivity, visual complexity, and wide variances among interpreters, human professionals ability to interpret images is relatively restricted. Consequently, early cancer detection using image processing techniques is possible. Image segmentation and the extraction of radiomic features from CT scans are the two main components of image processing. Deep learning is currently offering innovative, highly authentic solutions for medical imaging and is recognized as a crucial technique for upcoming applications in the health sector as a result of its success in other real-world applications. A a deep learning model along with image processing techniques is proposed for detecting lung cancer. SPIE-AAPM Lung CT Challenge dataset is used for this study. The proposed Deep Learning Model based Lung Cancer Detection (DLMLCD) achieves an accuracy of 99% on evaluation of the above dataset.
Prasenjit Dhar, K. Suganya Devi, Satish Kumar Satti, and P. Srinivasan
Springer Science and Business Media LLC
Satish Kumar Satti, K Suganya Devi, K Sekar, Prasenjit Dhar, and P Srinivasan
IEEE
Automatic traffic sign detection is a research topic for various applications, including driver assistance, inventory management, and autonomous driving. The visual appearance of traffic signs varies significantly from country to country, making classification systems more challenging to implement. In this work classification of traffic signs in the Indian roadways is considered. But there is currently no publicly available dataset for Indian traffic signs. This paper proposes the Indian Cautionary Traffic Sign (ICTS) data-set with 19,775 samples collected under challenging environmental conditions. The dataset is preprocessed using image processing techniques before being labeled into 40 categories. Then, to classify ICTS traffic signs, deep learning classification models such as LeNET, Vgg-16, AlexNet, and ResNet are used. The performance of all of these trained models is compared and analyzed on the proposed dataset and the GTSRB-benchmark dataset in terms of specificity, precision, recall, F-measure, and accuracy. The proposed dataset is available at the IEEE data port repository (http://dx.doi.org/10.21227/yy4h-rc98).
K. Sekar, K. Suganya Devi, and P. Srinivasan
Springer Science and Business Media LLC
Satish Kumar Satti, K. Suganya Devi, Prasenjit Dhar, and P. Srinivasan
Elsevier BV
Abstract Road boundary lanes are one of the serious causes of road accidents and it affects the driver and people’s safety. Detecting road boundary lanes is a challenging task for both computer vision and machine learning approaches. In recent years many machine learning algorithms have been deploying but they failed to produce high efficiency and accuracy. This paper presents a novel approach to alert the driver when the car leaps beyond the Road boundary lanes by employing machine learning techniques to avoid road mishaps and ensuring driving safety. Performance is assessed through the generation of experimental results on the dataset. When compared with state-of-the-art lane detection techniques, the proposed technique produced high precision and high efficiency.
K. Suganya Devi, G. Arutperumjothi, and P. Srinivasan
Springer Singapore
Prasenjit Dhar, K. Suganya Devi, Satish Kumar Satti, and P. Srinivasan
Springer Singapore
Hemanth Kumar Vasireddi and K. Suganya Devi
Springer Singapore
Badal Soni, K. Suganya Devi, and Angshuman Bora
Springer International Publishing
The proposed work solely focuses on transforming the manual task of pathologists in classifying a test mutation to a task automatically done by a machine. We collected the dataset from a Kaggle competition which distributes the three features, Gene, Variation and Text into nine different classes. These classes are provided by genomic researchers which state whether a particular mutation is a driver (cancer causing mutation) or a passenger (neutral mutation). Our model was able to perform the labour intensive work of classification, thus saving time, diminishing possibility of human error and risks of wrong analysis. As the score was not too high, we have given a probabilistic output and hence our model is interpretable. Researchers need to analyse only around two classes with highest probability for classification ignoring all other classes. We have used different encoding and embedding techniques to convert text entry to numerical form, after which they are fed as input to different classifiers. Logistic Regression classifier with Term Frequency—Inverse Document Frequency encoding technique fetched the maximum accuracy, i.e. 67.27%. Attention had been given to enhance this accuracy with Word2Vec and Doc2Vec embedding, but it only decreased due to some issues with our dataset as discussed later. However, with a better dataset, our model can ensure better accuracy.
N. V. Raja Reddy Goluguri, K. Suganya Devi, and Nagesh Vadaparthi
Springer Science and Business Media LLC
Growth in consumption of Oryza sativa (rice) has led the farmers across Asian countries to cultivate Oryza sativa , with an impact of 2.5 percent increase in the cultivation of the crop every year. Along with the growth in Oryza sativa cultivation, there are new challenges that are faced by the farmers in terms of diseases. The absence of information to recognize what sort of infection the plant is influenced with during the harvest cycle drives the farmers over the globe to lose 37 percent of the production. Involving technology to identify these diseases during the harvest cycle will help the farmers to get benefitted by attaining better yields. Deep learning being a latest technology playing a vital role in helping human in many aspects. A thorough review of the research papers on the various classifiers used in the identification of Oryza sativa diseases was carried out and the survey was tabulated and presented.
K. Sekar, K. Suganya Devi, P. Srinivasan and V. Senthilkumar
The deep learning-based compressive Sensing (CS) has shown substantial improved performance and in run-time reduction with signal sampling and reconstruction. In most cases, moreover, these techniques suffer from disrupting artefacts or high-frequency contents at low sampling ratios. Similarly, this occurs in the multi-resolution sampling method, which further collects more components with lower frequencies. A promising innovation combining CS with convolutionary neural network has eliminated the sparsity constraint yet recovery persists slow. We propose a Deep wavelet based compressive sensing with multi-resolution framework provides better improvement in reconstruction as well as run time. The proposed model demonstrates outstanding quality on test functions over previous approaches.