Rabindra Khadka

• Sengupta, Poushali; Khadka, Rabindra; Maharjan, Sabita; Eliassen, Frank; Pandey, Shashi Raj & Zhang, Yan [Vis alle 8 forfattere av denne artikkelen] (2026). Explainability of Complex AI Models with Correlation Impact Ratio. IEEE Transactions on Artificial Intelligence (TAI). s. 14–14. doi: 10.1109/tai.2026.3732283. Fulltekst i vitenarkiv
• Khadka, Rabindra; Sengupta, Poushali; Lind, Pedro & Yazidi, Anis (2025). Quantization of Vision Transformer-Based Model for Real-Time EEG Classification. I Azab, Abdulrahman & Malki, Tomasz (Red.), Nordic e-Infrastructure Tomorrow 6th Nordic e-Infrastructure Collaboration Conference NeIC 2024, Tallinn, Estonia, May 27–29, 2024 Proceedings. Springer Nature. ISSN 9783031862403. s. 17–27. doi: 10.1007/978-3-031-86240-3_2. Fulltekst i vitenarkiv Vis sammendrag

  Abstract Electroencephalography (EEG) is a non-invasive and cost-effective tool for capturing brain signals. However, existing approaches based on deep learning models for classifying EEG signals are primarily trained on large EEG datasets and require extensive computational resources. This poses significant challenges for real-time processing, particularly in resource-constrained environments such as wearable devices and edge computing. Our work addresses these challenges by exploring quantization techniques to optimize a vision transformer model for real-time EEG classification. We apply post-training quantization to the proposed Vision Transformer (ViT) model, termed the Brain Signal Vision Transformer (BSVT), to reduce computational overhead while maintaining high performance. The BSVT model, trained on the ex3 HPC system, efficiently handles large datasets. We comprehensively analyze the trade-offs between model accuracy and computational efficiency. Specifically, we demonstrate the use of dynamic quantization on a ViT-based model for EEG classification, achieving a 2.49x reduction in model size and a 1.46x improvement in inference speed. These findings underscore the potential of quantization techniques and distributed computing to revolutionize the adoption of AI in healthcare.

• Khadka, Rabindra; Lind, Pedro; Yazidi, Anis & Belhadi, Asma (2025). DREAMS: A python framework for training deep learning models on EEG data with model card reporting for medical applications. SoftwareX. 30. doi: 10.1016/j.softx.2025.102140. Fulltekst i vitenarkiv
• Khan, Haroon; Khadka, Rabindra; Sultan, Malik Shahid; Yazidi, Anis; Ombao, Hernando & Mirtaheri, Peyman (2024). Unleashing the potential of fNIRS with machine learning: classification of fine anatomical movements to empower future brain-computer interface. Frontiers in Human Neuroscience. 18. doi: 10.3389/fnhum.2024.1354143. Fulltekst i vitenarkiv Vis sammendrag

  In this study, we explore the potential of using functional near-infrared spectroscopy (fNIRS) signals in conjunction with modern machine-learning techniques to classify specific anatomical movements to increase the number of control commands for a possible fNIRS-based brain-computer interface (BCI) applications. The study focuses on novel individual finger-tapping, a well-known task in fNIRS and fMRI studies, but limited to left/right or few fingers. Twenty-four right-handed participants performed the individual finger-tapping task. Data were recorded by using sixteen sources and detectors placed over the motor cortex according to the 10-10 international system. The event's average oxygenated Δ HbO and deoxygenated Δ HbR hemoglobin data were utilized as features to assess the performance of diverse machine learning (ML) models in a challenging multi-class classification setting. These methods include LDA, QDA, MNLR, XGBoost, and RF. A new DL-based model named “Hemo-Net” has been proposed which consists of multiple parallel convolution layers with different filters to extract the features. This paper aims to explore the efficacy of using fNRIS along with ML/DL methods in a multi-class classification task. Complex models like RF, XGBoost, and Hemo-Net produce relatively higher test set accuracy when compared to LDA, MNLR, and QDA. Hemo-Net has depicted a superior performance achieving the highest test set accuracy of 76%, however, in this work, we do not aim at improving the accuracies of models rather we are interested in exploring if fNIRS has the neural signatures to help modern ML/DL methods in multi-class classification which can lead to applications like brain-computer interfaces. Multi-class classification of fine anatomical movements, such as individual finger movements, is difficult to classify with fNIRS data. Traditional ML models like MNLR and LDA show inferior performance compared to the ensemble-based methods of RF and XGBoost. DL-based method Hemo-Net outperforms all methods evaluated in this study and demonstrates a promising future for fNIRS-based BCI applications.

• Nguyen, Thu; Khadka, Rabindra; Phan, Nhan; Yazidi, Anis; Halvorsen, Pål & Hammer, Hugo Lewi [Vis alle 7 forfattere av denne artikkelen] (2022). Combining datasets to improve model fitting, 2022 International Joint Conference on Neural Networks (IJCNN). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781728186719. doi: 10.1109/IJCNN54540.2023.10191273. Fulltekst i vitenarkiv
• Abolghasemi, Roza; Khadka, Rabindra; Lind, Pedro; Engelstad, Paal E.; Viedma, Enrique Herrera & Yazidi, Anis (2022). Predicting missing pairwise preferences from similarity features in group decision making. Knowledge-Based Systems. ISSN 0950-7051. 256(256). doi: 10.1016/j.knosys.2022.109860. Fulltekst i vitenarkiv Vis sammendrag

  In group decision-making (GDM), fuzzy preference relations (FPRs) refer to pairwise preferences in the form of a matrix. Within the field of GDM, the problem of estimating missing values is of utmost importance, since many experts provide incomplete preferences. In this paper, we propose a new method called the entropy-based method for estimating the missing values in the FPR. We compared the accuracy of our algorithm for predicting the missing values with the best candidate algorithm from state of the art achievements. In the proposed entropy-based method, we took advantage of pairwise preferences to achieve good results by storing extra information compared to single rating scores, for example, a pairwise comparison of alternatives vs. the alternative’s score from one to five stars. The entropy-based method maps the prediction problem into a matrix factorization problem, and thus the solution for the matrix factorization can be expressed in the form of latent expert features and latent alternative features. Thus, the entropy-based method embeds alternatives and experts in the same latent feature space. By virtue of this embedding, another novelty of our approach is to use the similarity of experts, as well as the similarity between alternatives, to infer the missing values even when only minimal data are available for some alternatives from some experts. Note that current approaches may fail to provide any output in such cases. Apart from estimating missing values, another salient contribution of this paper is to use the proposed entropy-based method to rank the alternatives. It is worth mentioning that ranking alternatives have many possible applications in GDM, especially in group recommendation systems (GRS).

• Khadka, Rabindra; Jha, Debesh; Hicks, Steven; Thambawita, Vajira L B; Riegler, Michael & Ali, Sharib [Vis alle 7 forfattere av denne artikkelen] (2022). Meta-learning with implicit gradients in a few-shot setting for medical image segmentation. Computers in Biology and Medicine. ISSN 0010-4825. 143(143). doi: 10.1016/j.compbiomed.2022.105227. Fulltekst i vitenarkiv Vis sammendrag

  Widely used traditional supervised deep learning methods require a large number of training samples but often fail to generalize on unseen datasets. Therefore, a more general application of any trained model is quite limited for medical imaging for clinical practice. Using separately trained models for each unique lesion category or a unique patient population will require sufficiently large curated datasets, which is not practical to use in a real-world clinical set-up. Few-shot learning approaches can not only minimize the need for an enormous number of reliable ground truth labels that are labour-intensive and expensive, but can also be used to model on a dataset coming from a new population. To this end, we propose to exploit an optimization-based implicit model agnostic meta-learning (iMAML) algorithm under few-shot settings for medical image segmentation. Our approach can leverage the learned weights from diverse but small training samples to perform analysis on unseen datasets with high accuracy. We show that, unlike classical few-shot learning approaches, our method improves generalization capability. To our knowledge, this is the first work that exploits iMAML for medical image segmentation and explores the strength of the model on scenarios such as meta-training on unique and mixed instances of lesion datasets. Our quantitative results on publicly available skin and polyp datasets show that the proposed method outperforms the naive supervised baseline model and two recent few-shot segmentation approaches by large margins. In addition, our iMAML approach shows an improvement of 2%–4% in dice score compared to its counterpart MAML for most experiments.

Se alle arbeider i NVA

• Khadka, Rabindra; Lind, Pedro; Mello, Gustavo Borges Moreno E; Riegler, Michael Alexander & Yazidi, Anis (2024). Inducing Inductive Bias in Vision Transformer for EEG Classification. Fulltekst i vitenarkiv
• Haraldsen, Ira Hebold; Hatlestad-Hall, Christoffer; Marra, Camillo; Renvall, Hanna; Maestú, Fernando & Acosta-Hernández, Jorge [Vis alle 69 forfattere av denne artikkelen] (2023). Intelligent digital tools for screening of brain connectivity and dementia risk estimation in people affected by mild cognitive impairment: the AI-Mind clinical study protocol. Frontiers in Neurorobotics. 17. doi: 10.3389/fnbot.2023.1289406. Fulltekst i vitenarkiv Vis sammendrag

  More than 10 million Europeans show signs of mild cognitive impairment (MCI), a transitional stage between normal brain aging and dementia stage memory disorder. The path MCI takes can be divergent; while some maintain stability or even revert to cognitive norms, alarmingly, up to half of the cases progress to dementia within 5 years. Current diagnostic practice lacks the necessary screening tools to identify those at risk of progression. The European patient experience often involves a long journey from the initial signs of MCI to the eventual diagnosis of dementia. The trajectory is far from ideal. Here, we introduce the AI-Mind project, a pioneering initiative with an innovative approach to early risk assessment through the implementation of advanced artificial intelligence (AI) on multimodal data. The cutting-edge AI-based tools developed in the project aim not only to accelerate the diagnostic process but also to deliver highly accurate predictions regarding an individual’s risk of developing dementia when prevention and intervention may still be possible. AI-Mind is a European Research and Innovation Action (RIA H2020-SC1-BHC-06-2020, No. 964220) financed between 2021 and 2026. First, the AI-Mind Connector identifies dysfunctional brain networks based on high-density magneto- and electroencephalography (M/EEG) recordings. Second, the AI-Mind Predictor predicts dementia risk using data from the Connector, enriched with computerized cognitive tests, genetic and protein biomarkers, as well as sociodemographic and clinical variables. AIMind is integrated within a network of major European initiatives, including The Virtual Brain, The Virtual Epileptic Patient, and EBRAINS AISBL service for sensitive data, HealthDataCloud, where big patient data are generated for advancing digital and virtual twin technology development. AI-Mind’s innovation lies not only in its early prediction of dementia risk, but it also enables a virtual laboratory scenario for hypothesis-driven personalized intervention research. This article introduces the background of the AI-Mind project and its clinical study protocol, setting the stage for future scientific contribution.

• Khadka, Rabindra (2026). Structured Representations for EEG-Based Dementia Prediction: From Spatiotemporal Transformers to Multimodal Biomarker Regression. OsloMet - storbyuniversitetet. ISSN 9788283647624. Fulltekst i vitenarkiv Vis sammendrag

  Dementia poses a growing global health challenge, yet current diagnostic pathways depend on invasive and expensive biomarkers, such as cerebrospinal fluid assays and PET imaging, which limit scalability for early detection and longitudinal monitoring. Electroencephalography (EEG) offers a low-cost, non-invasive window into brain dynamics, but its clinical utility remains constrained by non-specificity, heterogeneous acquisition protocols, and limited interpretability. In this thesis, we address these limitations by advancing the concept of structure-aware representation learning, developing machine learning frameworks that preserve, discover, and translate the rich spatiotemporal and network-level structure inherent in EEG into clinically actionable biomarkers. Our work represents an early demonstration of continuous p-tau217 regression, a key molecular biomarker of Alzheimer’s disease, directly from EEG combined with clinical metadata. We begin by designing a transformer-based architecture with spatiotemporal inductive biases, the Brain Signal Vision Transformer (BSVT), that captures spatial correlations across electrodes and temporal dynamics within signals. We show that this structure-aware design improves data efficiency and downstream classification performance compared to standard ViT baselines. We then develop a selfsupervised foundation model (EEG-VJEPA) that learns transferable representations from large-scale unlabeled EEG, capturing physiologically meaningful structure and enabling robust performance in data-scarce clinical scenarios. To enhance interpretability, we extract microstate-based features that provide physiologically grounded biomarkers of cognitive decline and use an explainable framework to reveal network-level mechanisms associated with disease progression. Building on the representation learned with EEG-VJEPA, we propose a multimodal fusion framework that combines EEG embeddings with demographic, cognitive, and genetic metadata to predict plasma p-tau217 levels. Through a montageadaptive projection and a biomarker-aware modulation layer (BioFiLM), we demonstrate the first non-invasive approximation of molecular pathology from EEG (MAE = 2.55), outperforming unimodal baselines and highlighting the synergistic value of multimodal integration. Finally, we address key barriers to clinical deployment. We show that posttraining quantization compresses the transformer model (BSVT) by∼59% and accelerates inference by ∼47% with minimal accuracy loss, enabling real-time EEG analysis on resource-constrained edge devices. In parallel, we introduce DREAMS, an automated model-card framework that documents preprocessing, performance, uncertainty, and versioning, enhancing transparency, reproducibility, and regulatory readiness. Overall, our results demonstrate that embedding domain knowledge into deep learning architectures, leveraging self-supervision, integrating interpretable features, and fusing complementary data streams enable EEG to evolve from a purely functional marker into a non-invasive proxy for molecular pathology. This work provides a comprehensive methodological foundation for translating EEG into clinically actionable knowledge, paving the way for precision diagnostics and scalable biomarker discovery in neurodegenerative disease.

Se alle arbeider i NVA