Pankaj Pandey

Ruthwik Reddy Doodipala, Pankaj Pandey, Pratheek Eranki et al.

Self-supervised pretraining is promising for large-scale neuroimaging, yet the impact of region-aware masking and hybrid sequence modeling remains underexplored. In this work, we introduce Rhamba, a region-aware pretraining framework that integrates anatomically guided masking with hybrid Attention-Mamba architectures for resting state functional magnetic resonance imaging (fMRI) analysis. Models were pretrained on the ABIDE dataset using region-aligned patch embeddings and three masking strategies (Any, Majority, and Pure) with increasing spatial specificity. We evaluated four architectural variants: a Mamba only model, an Alternate architecture with interleaved Mamba and Attention blocks, and two hybrid encoder-decoder configurations (Attention-Mamba (AM) and Mamba-Attention (MA)). The pretrained models were fine-tuned on downstream classification tasks using the COBRE and ADHD-200 datasets for schizophrenia and attention-deficit/hyperactivity disorder discrimination. We employed Integrated Gradients, an explainable AI method, to identify the brain regions contributing to model predictions. Masking strategy strongly influenced reconstruction behavior, with reconstruction loss following a consistent ordering (Any > Majority > Pure). However, this trend did not directly translate into downstream performance, where differences were modest and dataset-dependent. The hybrid architecture with the MA configuration achieved the highest average AUROC across both datasets, and Rhamba outperformed state-of-the-art methods in comparative evaluation. Region-wise analysis showed that peak performance depends on the interaction between masking strategy and architecture rather than a single dominant configuration. Overall, Rhamba offers a flexible framework for balancing interpretability, scalability, and performance in large-scale fMRI representation learning.

Sudip Das, Pankaj Pandey, Krishna Prasad Miyapuram

There is abundant medical data on the internet, most of which are unlabeled. Traditional supervised learning algorithms are often limited by the amount of labeled data, especially in the medical domain, where labeling is costly in terms of human processing and specialized experts needed to label them. They are also prone to human error and biased as a select few expert annotators label them. These issues are mitigated by Self-supervision, where we generate pseudo-labels from unlabelled data by seeing the data itself. This paper presents various self-supervision strategies to enhance the performance of a time-series based Diffusion convolution recurrent neural network (DCRNN) model. The learned weights in the self-supervision pretraining phase can be transferred to the supervised training phase to boost the model's prediction capability. Our techniques are tested on an extension of a Diffusion Convolutional Recurrent Neural network (DCRNN) model, an RNN with graph diffusion convolutions, which models the spatiotemporal dependencies present in EEG signals. When the learned weights from the pretraining stage are transferred to a DCRNN model to determine whether an EEG time window has a characteristic seizure signal associated with it, our method yields an AUROC score $1.56\%$ than the current state-of-the-art models on the TUH EEG seizure corpus.

Ruthwik Reddy Doodipala, Pankaj Pandey, Carolina Torres Rojas et al.

The emergence of foundation models in neuroimaging is driven by the increasing availability of large-scale and heterogeneous brain imaging datasets. Recent advances in self-supervised learning, particularly reconstruction-based objectives, have demonstrated strong potential for pretraining models that generalize effectively across diverse downstream functional MRI (fMRI) tasks. In this study, we explore region-aware reconstruction strategies for a foundation model in resting-state fMRI, moving beyond approaches that rely on random region masking. Specifically, we introduce an ROI-guided masking strategy using the Automated Anatomical Labelling Atlas (AAL3), applied directly to full 4D fMRI volumes to selectively mask semantically coherent brain regions during self-supervised pretraining. Using the ADHD-200 dataset comprising 973 subjects with resting-state fMRI scans, we show that our method achieves a 4.23% improvement in classification accuracy for distinguishing healthy controls from individuals diagnosed with ADHD, compared to conventional random masking. Region-level attribution analysis reveals that brain volumes within the limbic region and cerebellum contribute most significantly to reconstruction fidelity and model representation. Our results demonstrate that masking anatomical regions during model pretraining not only enhances interpretability but also yields more robust and discriminative representations. In future work, we plan to extend this approach by evaluating it on additional neuroimaging datasets, and developing new loss functions explicitly derived from region-aware reconstruction objectives. These directions aim to further improve the robustness and interpretability of foundation models for functional neuroimaging.

Behtom Adeli, John McLinden, Pankaj Pandey et al.

Activation functions are critical to the performance of deep neural networks, particularly in domains such as functional near-infrared spectroscopy (fNIRS), where nonlinearity, low signal-to-noise ratio (SNR), and signal variability poses significant challenges to model accuracy. However, the impact of activation functions on deep learning (DL) performance in the fNIRS domain remains underexplored and lacks systematic investigation in the current literature. This study evaluates a range of conventional and field-specific activation functions for fNIRS classification tasks using multiple deep learning architectures, including the domain-specific fNIRSNet, AbsoluteNet, MDNN, and shallowConvNet (as the baseline), all tested on a single dataset recorded during an auditory task. To ensure fair a comparison, all networks were trained and tested using standardized preprocessing and consistent training parameters. The results show that symmetrical activation functions such as Tanh and the Absolute value function Abs(x) can outperform commonly used functions like the Rectified Linear Unit (ReLU), depending on the architecture. Additionally, a focused analysis of the role of symmetry was conducted using a Modified Absolute Function (MAF), with results further supporting the effectiveness of symmetrical activation functions on performance gains. These findings underscore the importance of selecting proper activation functions that align with the signal characteristics of fNIRS data.

Behtom Adeli, John Mclinden, Pankaj Pandey et al.

In recent years, deep learning (DL) approaches have demonstrated promising results in decoding hemodynamic responses captured by functional near-infrared spectroscopy (fNIRS), particularly in the context of brain-computer interface (BCI) applications. This work introduces AbsoluteNet, a novel deep learning architecture designed to classify auditory event-related responses recorded using fNIRS. The proposed network is built upon principles of spatio-temporal convolution and customized activation functions. Our model was compared against several models, namely fNIRSNET, MDNN, DeepConvNet, and ShallowConvNet. The results showed that AbsoluteNet outperforms existing models, reaching 87.0% accuracy, 84.8% sensitivity, and 89.2% specificity in binary classification, surpassing fNIRSNET, the second-best model, by 3.8% in accuracy. These findings underscore the effectiveness of our proposed deep learning model in decoding hemodynamic responses related to auditory processing and highlight the importance of spatio-temporal feature aggregation and customized activation functions to better fit fNIRS dynamics.

Pankaj Pandey, Krishna Prasad Miyapuram

Several Convolutional Deep Learning models have been proposed to classify the cognitive states utilizing several neuro-imaging domains. These models have achieved significant results, but they are heavily designed with millions of parameters, which increases train and test time, making the model complex and less suitable for real-time analysis. This paper proposes a simple, lightweight CNN model to classify cognitive states from Electroencephalograph (EEG) recordings. We develop a novel pipeline to learn distinct cognitive representation consisting of two stages. The first stage is to generate the 2D spectral images from neural time series signals in a particular frequency band. Images are generated to preserve the relationship between the neighboring electrodes and the spectral property of the cognitive events. The second is to develop a time-efficient, computationally less loaded, and high-performing model. We design a network containing 4 blocks and major components include standard and depth-wise convolution for increasing the performance and followed by separable convolution to decrease the number of parameters which maintains the tradeoff between time and performance. We experiment on open access EEG meditation dataset comprising expert, nonexpert meditative, and control states. We compare performance with six commonly used machine learning classifiers and four state of the art deep learning models. We attain comparable performance utilizing less than 4\% of the parameters of other models. This model can be employed in a real-time computation environment such as neurofeedback.

Pankaj Pandey, Raunak Swarnkar, Shobhit Kakaria et al.

Neuromarketing aims to understand consumer behavior using neuroscience. Brain imaging tools such as EEG have been used to better understand consumer behavior that goes beyond self-report measures which can be a more accurate measure to understand how and why consumers prefer choosing one product over another. Previous studies have shown that consumer preferences can be effectively predicted by understanding changes in evoked responses as captured by EEG. However, understanding ordered preference of choices was not studied earlier. In this study, we try to decipher the evoked responses using EEG while participants were presented with naturalistic stimuli i.e. movie trailers. Using Machine Learning tech niques to mine the patterns in EEG signals, we predicted the movie rating with more than above-chance, 72% accuracy. Our research shows that neural correlates can be an effective predictor of consumer choices and can significantly enhance our understanding of consumer behavior.

Pankaj Pandey

Today business environment is highly dependent on complex technologies, and information is considered an important asset. Organizations are therefore required to protect their information infrastructure and follow an inclusive risk management approach. One way to achieve this is by aligning the information security investment decisions with respect to organizational strategy. A large number of information security investment models have are in the literature. These models are useful for optimal and cost-effective investments in information security. However, it is extremely challenging for a decision maker to select one or combination of several models to decide on investments in information security controls. We propose a framework to simplify the task of selecting information security investment model(s). The proposed framework follows the 'Context, Content, Process' approach, and this approach is useful in evaluation and prioritization of investments in information security controls in alignment with the overall organizational strategy.