Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
3 results
Search Results
Item VesselXnet - A lightweight and efficient encoder-decoder based model for Retinal Vessel Segmentation(Institute of Electrical and Electronics Engineers Inc., 2021) Narasimhadhan, A.V.; Putluru, S.P.R.; Merugu, V.R.One of the contemporary issues present in medical image segmentation is the segmentation of the Retinal blood vessels. This is because many diseases can be accurately identified from the vascular structure of the retina and hence can be treated early and diagnosed thoroughly. Manual segmentation is hectic, cumbersome, time consuming and also error-prone. Hence there is a need for automatic vessel segmentation which can be a better technological advancement in the medical field. Some of the segmentation methods which were proposed previously have problems of low segmentation accuracy, incomplete segmentation and a large model size. With the progress of deep learning and convolutional neural networks several U-net based architectures were extensively used for this task which offered reliable segmentation results. In this paper, we proposed a light weight U-net based architecture which provides comparable accuracy with much less total parameters. © 2021 IEEE.Item Fusing Conventional and Deep Learning Features for Hyperspectral Image Change Detection(Institute of Electrical and Electronics Engineers Inc., 2022) Bobate, N.; Yadav, P.P.; Narasimhadhan, A.V.In the field of remote sensing technology Change Detection (CD) is one of the major areas of research. Changes that have occurred on the earth's surface over time can be detected with this tool. Hyperspectral Image (HSI) data with high spectral resolution can help in identifying subtle changes than the typical multispectral image (MSI), and CD technology has benefitted immensely with the applications of HSI. Traditional CD techniques that used MSI as their input data are challenging to implement on HSI due to the high dimensionality of hyperspectral data. Furthermore, HSI data is affected by a lot of distortion and redundancy, contaminating the spectral-only information for CD purposes. CD accuracy can be improved by extracting the useful features of HSI. In Change Detection algorithms, the initial step is to extract features. Traditionally it is done using arithmetic operation, image transformation, and statistical methods. While some advanced strategies for extracting features are utilizing convolutional neural networks (CNNs) using the deep learning method. In this work, we aimed to integrate the conventional features with CNN extracted features to boost the overall ac-curacy of popular DL-based CD techniques. Spectral matching algorithms are used for extracting conventional features. In addition, appropriate changes are made to the recent deep learning architectures called Three-Directions Spectral-Spatial Convolution neural network (TDSSC) and General End-To-End Neural Network (GETNET), to fuse the conventional features. Farmland, River and USA data sets are used for experimentation. The proposed approach proves to be useful in improving the performance of DL-based CD techniques. © 2022 IEEE.Item LCSNet: Lightweight Caries Segmentation Network for the segmentation of dental caries using smartphone photographs(Elsevier B.V., 2025) Radha, R.C.; Raghavendra, B.S.; Hota, R.K.; Vijayalakshmi, K.R.; Patil, S.; Narasimhadhan, A.V.Dental caries is one of the major dental issues that is common among many individuals. It leads to tooth loss and affects the tooth root, creating a need to automatically detect dental caries to reduce treatment costs and prevent its consequences. The Lightweight Caries Segmentation Network (LCSNet) proposed in this study detects the location of dental caries by applying pixel-wise segmentation to dental photographs taken with various Android phones. LCSNet utilizes a Dual Multiscale Residual (DMR) block in both the encoder and decoder, adapts transfer learning through a pre-trained InceptionV3 model at the bottleneck layer, and incorporates a Squeeze and Excitation block in the skip connection, effectively extracting spatial information even from images where 95 % of the background and only 5 % represent the area of interest. A new dataset was developed by gathering oral photographs of dental caries from two hospitals, with advanced augmentation techniques applied. The LCSNet architecture demonstrated an accuracy of 97.36 %, precision of 73.1 %, recall of 70.2 %, an F1-Score of 71.14 %, and an Intersection-over-Union (IoU) of 56.8 %. Expert dentists confirmed that the LCSNet model proposed in this in vivo study accurately segments the position and texture of dental caries. Both qualitative and quantitative performance analyses, along with comparative analyses of efficiency and computational requirements, were conducted with other deep learning models. The proposed model outperforms existing deep learning models and shows significant potential for integration into a smartphone application-based oral disease detection system, potentially replacing some conventional clinically adapted methods. © 2025 The Authors.