Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
6 results
Search Results
Item Speech Summarization Using Prosodic Features and 1-D Convolutional Neural Network(Institute of Electrical and Electronics Engineers Inc., 2022) Chauhan, S.R.; Ambesange, S.; Koolagudi, S.G.In this work, we have presented a method for speech summarization of audiobooks without converting them into the transcript. The model used is the 1-D convolutional neural network. The audio is segmented into sentences based on the silence between two consecutive sentences. We have used acoustic features of the sentence audio as input to our model. The output of our model is binary, which tells us whether to include this sentence in our summary or not. Thus, we have converted the task of speech summarization into a classification task. Then we have concatenated the classified audio chunks into one summary. We have compared the generated summary against the manually done summary. For better insights, we have used a text summarizer as a reference to see what the summary should include. The transcript is used for only that; otherwise, our method is independent of the text. The results obtained show us a possibility of a language-independent audio summarizer that retains the audio quality since we have used the original audio in our summary. © 2022 IEEE.Item Simulating Federated Transfer Learning for Lung Segmentation using Modified UNet Model(Elsevier B.V., 2022) Ambesange, S.; Annappa, A.; Koolagudi, S.G.Lung segmentation helps doctors in analyzing and diagnosing lung diseases effectively. Covid -19 pandemic highlighted the need for such artificial intelligence (AI) model to segment Lung X-ray images and diagnose patient covid conditions, in a short time, which was not possible due to huge number of patient influx at hospitals with the limited radiologist to diagnose based on test report in short time. AI models developed to assist doctors to diagnose faster, faces another challenge of data privacy. Such AI Models, for better performance, need huge data collected from multiple hospitals/diagnostic centres across the globe into single place to train the AI models. Federated Learning (FL) framework, using transfer learning approach addresses these concerns as FL framework doesn't need data to be shared to outside hospital ecosystem, as AI model get trained on local system and AI model get trained on distributed data. FL with Transfer learning doesn't need the parallel training of the model at all participants nodes like other FL. Paper simulates Federated Transfer learning for Image segmentation using transfer learning technique with few participating nodes and each nodes having different size dataset. The proposed method also leverages other healthcare data available at local system to train the proposed model to overcome lack of more data. Paper uses pre-trained weights of U-net Segmentation Model trained for MRI image segmentation to lung segmentation model. Paper demonstrates using such similar healthcare data available at local system helps improving the performance of the model. The paper uses Explainable AI approach to explain the result. Using above three techniques, Lung segmentation AI model gets near perfect segmentation accuracy. © 2023 The Authors. Published by Elsevier B.V.Item Prostate Cancer Grading Using Multistage Deep Neural Networks(Springer Science and Business Media Deutschland GmbH, 2023) Bygari, R.; Rithesh, K.; Ambesange, S.; Koolagudi, S.G.Prostate cancer is the second most commonly occurring cancer in men with a high incidence to mortality ratio. Accurate prostate cancer grading is the foremost step in determining the precise treatment process for the patient in preventing mortality of the patient. Currently, the grading is carried out by pathologists, which has limitation of availability super specialist doctors across world to grade it at affordable price, and non-super specialist doctor grading is error prone. This paper evades the need for an expert pathologist by proposing a novel deep learning method for automatic screening of prostate images to detect and assign a grade severity of cancer based on the images. The explainability of classification model imbibed using gradient-weighted class activation mapping (GradCAM) visualization, which generate heatmap of image, which influenced the decision of the model. The proposed method has three stages with ensemble deep neural networks to grade the prostate cancer. Firstly, a UNet is used for the segmentation of the histopathological image. Subsequently, the segmented image is overlaid on the original image, which helps underscore the most critical regions determining the grade of cancer. Finally, the overlaid image is used by an ensemble model consisting of Xception, Resnet-50, EfficientNet-b7 to predict the final grade of the histopathological image. The dataset containing 10,000 histopathological images obtained from Karolinska and Radboud that are made publicly available through the Prostate Cancer Grade Assessment Challenge hosted in Kaggle is used for training and evaluation. This method achieves a classification accuracy of 92.38% and outperforms many state-of-the-art methods. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item FedCure: A Heterogeneity-Aware Personalized Federated Learning Framework for Intelligent Healthcare Applications in IoMT Environments(Institute of Electrical and Electronics Engineers Inc., 2024) Sachin, D.N.; Annappa, B.; Hegde, S.; Abhijit, C.S.; Ambesange, S.The advent of the Internet of Medical Things (IoMT) devices has led to a healthcare revolution, introducing a new era of smart applications driven by Artificial Intelligence (AI). These advanced technologies have greatly influenced the healthcare industry and have played a crucial role in enhancing the quality of life globally. Federated Learning (FL) has become popular as a technique to create models that can be shared universally using the vast datasets collected from IoMT devices while maintaining data privacy. However, the complex variations in IoMT environments, including diverse devices, data characteristics, and model complexities, create challenges for the straightforward application of traditional FL methods. Consequently, it is not well-suited for deployment in such contexts. This paper introduces FedCure, a personalized FL framework tailored for intelligent IoMT-based healthcare applications operating within a cloud-edge architecture. FedCure is adept at addressing the challenges within IoMT environments by employing personalized FL techniques that can effectively mitigate the impact of heterogeneity. Furthermore, the integration of edge computing technology enhances processing speed and minimizes latency in intelligent IoMT applications. Lastly, this research showcases several case studies encompassing IoMT-based applications, such as Eye Retinopathy Detection, Diabetes Monitoring, Maternal Health, Remote Health Monitoring, and Human Activity Recognition. These case studies provide a means to assess the effectiveness of the proposed FedCure framework and showcase exceptional performance with accuracy and minimal communication overhead, especially in addressing the challenges posed by heterogeneity. © 2013 IEEE.Item Federated learning for digital healthcare: concepts, applications, frameworks, and challenges(Springer, 2024) Sachin, D.N.; Annappa, B.; Ambesange, S.Various hospitals have adopted digital technologies in the healthcare sector for various healthcare-related applications. Due to the effect of the Covid-19 pandemic, digital transformation has taken place in many domains, especially in the healthcare domain; it has streamlined various healthcare activities. With the advancement in technology concept of telemedicine evolved over the years and led to personalized healthcare and drug discovery. The use of machine learning (ML) technique in healthcare enables healthcare professionals to make a more accurate and early diagnosis. Training these ML models requires a massive amount of data, including patients’ personal data, that need to be protected from unethical use. Sharing these data to train ML models may violate data privacy. A distributed ML paradigm called federated learning (FL) has allowed different medical research institutions, hospitals, and healthcare devices to train ML models without sharing raw data. This survey paper overviews existing research work on FL-related use cases and applications. This paper also discusses the state-of-the-art tools and techniques available for FL research, current shortcomings, and future challenges in using FL in healthcare. © The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature 2024.Item Smart client selection strategies for enhanced federated learning in digital healthcare applications(Springer, 2025) Sachin, D.N.; Annappa, B.; Ambesange, S.Federated Learning (FL) trains AI models in healthcare without sharing patient data. FL computes client models locally and combines them to create a global model. However, involving all clients is impractical due to resource limitations. Random selection of a subset of clients in each FL round can pose challenges for resource-limited devices, leading to longer processing times and potential training failures. To tackle these obstacles, this research proposes a novel strategy for FL that treats each training round as a client selection process to improve the efficiency and effectiveness of FL in healthcare applications, where data privacy is paramount. The approach begins by calculating the uncertainty value for each client, which quantifies the contribution of the client’s data to the overall model. Clients are then ranked based on their uncertainty values, and those with higher loss values are given a higher probability of participating in the training process. The experimental outcomes clearly show that the proposed strategy effectively makes 1.3x training faster, and 30% lowers communication expenses, conserves computational resources, and enhances model performance when contrasted with random client selection. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.