Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
Search Results
Item Improving the performance of multi-stage HER2 breast cancer detection in hematoxylin-eosin images based on ensemble deep learning(Elsevier Ltd, 2025) Pateel, G.P.; Senapati, K.; Pandey, A.K.Background: Breast cancer is the most frequently diagnosed cancer among women worldwide, and histopathology is the gold standard in diagnosing the disease. Hematoxylin and Eosin (HE) staining, routinely employed to observe the overall tissue structure, is an affordable and commonly practiced cancer diagnosis. In contrast, Immunohistochemistry (IHC), which detects the increased presence of particular antigens linked to the mutation, can require multiple tests to conduct and is relatively costly. Generally, in computer-aided diagnosis, the conventional methods rely on a single network to extract features. However, these methods have significant limitations and fail to generalize. Methods: In this study, we propose an automated novel weighted average algorithm called HER2-ETNET, which ensembles the chosen three pre-trained deep learning models, DenseNet 201, GoogLeNet, and ResNet-50, to classify breast histopathology HE images into multi-class Human Epidermal Growth Factor Receptor-2 (HER2) status (HER2-0+, HER2-1+, HER2-2+, HER2-3+). The proposed method has the potential to bypass the IHC laboratory test. In this study, we form a weight matrix by fusing together, the scores of False Positive Rate (FPR) and False Negative Rate (FNR) of both training and validation sets, and the computed weights are assigned to the three base learners. This is in contrast to the previous works, in which the weights were generally assigned empirically to the chosen deep learning models, which might be erroneous. Result: The proposed approach is evaluated on the unseen test set, and it achieves accuracy, precision, recall and AUC of 97.44%, 97.32%, 97.39%, and 99.75% respectively. Conclusion: The proposed framework outperforms all the existing methods on the same dataset and is proven to be the reliable method in detecting the HER2 status (HER2-0+, HER2-1+, HER2-2+, HER2-3+) from HE images. This also proves that, HE stained images contain adequate information for efficiently detecting the HER2 status in breast cancer. © 2024 Elsevier LtdItem BENN: Balanced Ensemble Neural Network for Handling Class Imbalance in Big Data(John Wiley and Sons Inc, 2025) Sneha, S.H.; Annappa, B.; Pariserum Perumal, S.P.Class imbalance is a critical challenge in big data analytics, often leading to biased predictive models. This imbalance can lead to biased models that perform well on the majority class but poorly on the minority class. Many machine learning models tend to be biased towards the majority class because they aim to minimise overall error, often leading to poor performance on the minority class. This paper presents the balanced ensemble neural network, a novel solution to effectively address class imbalance in big data. Balanced ensemble neural network combines the robust capabilities of neural networks with the power of ensemble learning, incorporating class balancing strategies to ensure fair representation of minority classes. The methodology involves integrating multiple neural networks, each trained on balanced subsets of data using techniques like Synthetic Minority Over-sampling Technique and Random Undersampling. This integration aims to leverage the strengths of individual networks while reducing their inherent biases. Our extensive experiments across various datasets reveal that BENN achieves an AUC-ROC score of 0.94, surpassing other models such as random forest (0.88), support vector (0.84) and single neural net (0.80). It was also observed that BENN's performance is better compared to traditional neural network models and standard ensemble methods in key metrics like accuracy, precision, recall, F1-score and AUC-ROC. The results specifically highlight BENN's effectiveness in accurately classifying instances of minority classes, a notable challenge in many existing models. These findings underscore BENN's potential as a substantial advancement in handling class imbalance within big data environments, offering a promising direction for future research and application in machine learning. © 2024 John Wiley & Sons Ltd.Item Empirical and machine learning-based approaches to identify rainfall thresholds for landslide prediction: a case study of Kerala, India(Springer Nature, 2025) Menon, V.; Kolathayar, S.Kerala, a state in India, experiences one of the highest incidences of rainfall-induced landslides. Historical data has been collected and analyzed to devise thresholds for the early detection of landslides. Two empirical approaches based on the relationships between rainfall intensity and duration, as well as cumulative rainfall and duration, have been utilized to identify early warning thresholds for landslides. Five machine learning-based approaches were employed to determine these thresholds. Among the classifiers tested, the K-Nearest Neighbour (KNN) classifier with K=5 demonstrated the highest prediction accuracy compared to other methods in the study.; For the safe and resilient development of cities, disaster risk reduction plays a crucial role, aligning with sustainable development goal 11 of the United Nations. Supporting this objective, the present study developed a machine learning (ML) classifier-based threshold model to determine rainfall thresholds for predicting impending landslides in Kerala, India, using historical data. Using a dataset of 64 rainfall-induced landslide events recorded since the year 2000, rainfall data were collected up to 15 days prior to each landslide to support empirical analysis of intensity-duration and event rainfall-duration thresholds. In cases where exact rainfall durations were unavailable, classification machine learning (ML) models, including K-nearest neighbours (KNN), random forest (RF), gradient boosting machine (GBM), support vector machine (SVM), and logistic regression, were used to determine threshold reliability. Among these, the KNN model with 5 Neighbours achieved the highest performance, with an ROC-AUC of 0.9 and an accuracy of 82%. This model, saved as a pickle file, serves as a core filter in the development of a landslide early warning system. This paper presents the model development and performance comparisons, contributing to a practical, community-centred solution for landslide disaster resilience in Kerala. © The Author(s) 2025.; © The Author(s) 2025.Item Improving Vertebral Fracture Detection in C-Spine CT Images Using Bayesian Probability-Based Ensemble Learning(Multidisciplinary Digital Publishing Institute (MDPI), 2025) Pandey, A.K.; Senapati, K.; Argyros, I.K.; Pateel, G.P.Vertebral fracture (VF) may induce spinal cord injury that can lead to serious consequences which eventually may paralyze the entire or some parts of the body depending on the location and severity of the injury. Diagnosis of VFs is crucial at the initial stage, which may be challenging because of the subtle features, noise, and homogeneity present in the computed tomography (CT) images. In this study, Wide ResNet-40, DenseNet-121, and EfficientNet-B7 are chosen, fine-tuned, and used as base models, and a Bayesian-based probabilistic ensemble learning method is proposed for fracture detection in cervical spine CT images. The proposed method considers the prediction’s uncertainty of the base models and combines the predictions obtained from them, to improve the overall performance significantly. This method assigns weights to the base learners, based on their performance and confidence about the prediction. To increase the robustness of the proposed model, custom data augmentation techniques are performed in the preprocessing step. This work utilizes 15,123 CT images from the RSNA-2022 C-spine fracture detection challenge and demonstrates superior performance compared to the individual base learners, and the other existing conventional ensemble methods. The proposed model also outperforms the best state-of-the-art (SOTA) model by 1.62%, 0.51%, and 1.29% in terms of accuracy, specificity, and sensitivity, respectively; furthermore, the AUC score of the best SOTA model is lagging by 5%. The overall accuracy, specificity, sensitivity, and F1-score of the proposed model are 94.62%, 93.51%, 95.29%, and 93.16%, respectively. © 2025 by the authors.Item Federated learning approach for human activity recognition in online examination environment(Springer, 2025) Ramu, S.; Guddeti, R.M.R.; Mohan, B.R.In recent years, online exams have become a key method for assessing students’ knowledge and skills. However, with the rise of e-learning, conducting these exams has introduced new challenges, especially due to the increasing tendency of students to engage in cheating during online assessments. To address this, student activities during online exams are monitored to detect behaviors that indicate cheating through Human Activity Recognition (HAR). HAR is a system capable of recognizing various human activities based on observational data. This study focuses on detecting student behavior during online exams, categorizing normal activities as non-cheating and abnormal activities as potential cheating or malpractice. For this purpose, a federated learning architecture was utilized to process online exam data. In this approach, we implemented federated models, including Federated-ResNet50, Federated-DenseNet121, Federated-VGG16, and Federated-CNN, to classify student activities. A OEP dataset is utilized in this work comprising of various activities such as using mobile devices, copying from notes, abnormal head gaze and normal. Model performance for classification was evaluated using accuracy, precision, recall and F1-Score metrics. The results were compared across the federated models, namely Federated-ResNet50, Federated-DenseNet121, Federated-VGG16, and Federated-CNN. Among these, Federated-ResNet50 performed the best, achieving an accuracy of 91.28%. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.