About

Ankit B. Patel is a faculty member at Rice University, specializing in Artificial Intelligence and Machine Learning within the School of Engineering and Computing. His research focuses on advancing the fields of AI and ML, contributing to the academic community through his work at Rice CS, Rice ECE, and Rice STAT. He is actively involved in teaching and research, fostering developments in data science and related areas. His contact information includes ankit.b.patel@rice.edu, and he is based at the Rice University campus in Houston, Texas.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Applied mathematics
• Theoretical computer science
• Mathematics
• Algorithm

Selected publications

• Artifact Identification and Mitigation Strategies for Longitudinal Neural Data Collection Onboard the Medtronic Percept DBS Device ^*

  medRxiv · 2025-07-23 · 1 citations

  preprintOpen access

  Recent advances in deep brain stimulation (DBS) devices have enabled the ability to capture continuous neural recordings concurrently with stimulation therapy in the background of everyday life. These recordings provide the opportunity to investigate neural biomarkers of various behaviors or clinical status. However, they are susceptible to artifacts that can obscure and limit our ability to interpret neural signals. In a cohort of 23 patients who underwent DBS for obsessive-compulsive disorder (OCD) with the Medtronic Percept device, we identified an artifact in longitudinal neural recordings that occurs when the detected voltage exceeds the device's maximum sensing capabilities. When such an event occurs, the device inserts a flag value in the neural power stream. We found that overvoltage events are significantly more common in patients implanted with legacy Medtronic 3387 leads than those with newer Medtronic SenSight leads. We demonstrate a best practice, principled strategy for correcting samples affected by overvoltage events to preserve the ability to analyze the data. Finally, in a subset of patients who wore an Oura Ring concurrently with neural recordings (N=14), we found that overvoltage events were more likely during physical activity, suggesting that movement artifacts may elevate low-frequency power regardless of lead model.

  PublisherOA PDFDOI

• Fine-Tuned Mistral Model for Multi-Agent Mental Health Counseling System

  2025-03-28

  article1st authorCorresponding

  In today’s digital world, mental health issues such as stress, depression, and anxiety are often expressed through social media. We propose a multi-agent system utilizing the fine-tuned Mistral-7B model to detect and provide personalized counseling for mental health concerns. This paper discusses the development and deployment of a multi-agent system consisting of diagnostic and counseling agents. By fine-tuning Mistral with the Interpretable Mental Health Instruction (IMHI) dataset, the system offers personalized mental health support, improving scalability and accessibility to mental health services.

  PublisherDOI

• Interpreting convolutional neural networks' low-dimensional approximation to quantum spin systems

  Physical Review Research · 2025-01-23

  articleOpen accessSenior author

  Convolutional neural networks (CNNs) have been employed along with variational Monte Carlo methods for finding the ground state of quantum many-body spin systems with great success. However, it remains uncertain how CNNs, with a model complexity that scales at most linearly with the number of particles, solve the “curse of dimensionality” and efficiently represent wavefunctions in exponentially large Hilbert spaces. In this work, we use methodologies from information theory, group theory and machine learning, to elucidate how CNN captures relevant physics of quantum systems. We connect CNNs to a class of restricted maximum entropy (MaxEnt) and entangled plaquette correlator product state (EP-CPS) models that approximate symmetry constrained classical correlations between subsystems. For the final part of the puzzle, inspired by similar analyses for matrix product states and tensor networks, we show that the CNNs rely on the spectrum of each subsystem's entanglement Hamiltonians as captured by the size of the convolutional filter. All put together, these allow CNNs to simulate exponential quantum wave functions using a model that scales at most linear in system size as well as provide clues into when CNNs might fail to simulate Hamiltonians. We incorporate our insights into a new training algorithm and demonstrate its improved efficiency, accuracy, and robustness. Finally, we use regression analysis to show how the CNNs solutions can be used to identify salient physical features of the system that are the most relevant to an efficient approximation. Our integrated approach can be extended to similarly analyzing other neural network architectures and quantum spin systems.

  PublisherOA PDFDOI

• Predicting dementia risk using neuroimaging and cognitive assessment

  Journal of Aging Research and Lifestyle · 2025-01-01

  articleOpen access

  Introduction: Dementia affects over 55 million people globally, with numbers expected to double in the coming decades. Early detection is critical, yet traditional risk assessments relying on age, family history, and basic cognitive tests often fall short. This study explores whether combining structural brain imaging with brief cognitive assessments can more accurately predict dementia risk. Method: Using data from 312 older adults enrolled in the KU Alzheimer's Disease Center cohort, researchers evaluated two modeling approaches: one based on a single clinic visit and another using longitudinal data across multiple visits. Participants underwent cognitive testing and MRI scans, including measures of hippocampal volume, gray matter, and Alzheimer's disease signature regions. Depressive symptoms were also assessed using the Geriatric Depression Scale (GDS). Results: Results showed that models incorporating neuroimaging significantly outperformed those using demographics or cognitive scores alone. The best-performing model combined imaging and cognitive data, achieving 77.6% accuracy in predicting dementia status. Longitudinal models further improved prediction by capturing changes over time, with imaging features contributing most to explained variance. Key predictors included reduced hippocampal volume, lower gray matter, and higher GDS scores. These findings align with known patterns of neurodegeneration and suggest that depression may interact with brain changes to influence dementia risk. Conclusion: Importantly, the study demonstrates that a compact, multimodal approaching standard MRI scans with brief cognitive tests-can generate individualized risk profiles suitable for clinical use. This method offers a scalable path to early intervention, trial enrollment, and personalized care planning. Future work will focus on validating these models in more diverse populations and integrating fluid biomarkers to enhance precision. Ultimately, this research supports the development of practical tools for forecasting dementia risk and advancing preventive strategies in aging populations.

  PublisherDOI

• Structural Neuroimaging Markers of Dementia: Insights from ROC Curve Analysis

  Research Square · 2025-09-17

  preprintOpen access
  PublisherOA PDFDOI

• A Deep Learning Framework for Quantifying Dynamic Self-Organization in <i>Myxococcus xanthus</i>

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-11-09

  preprintOpen access

  Abstract Under starvation, Myxococcus xanthus bacteria initiate a multicellular developmental program in which cells move to form fruiting bodies and differentiate into distinct cell types. Many genes affecting this process have been identified, and it is assumed that perturbing genes within the same pathway induces similar changes in the phenotype, although those changes may be subtle or obscured by pleiotropic effects. However, these pathways cannot be systematically mapped, as there are no systematic methods for quantifying phenotype similarity. Here, we applied deep learning techniques to quantify the phenotype patterns and self-organization dynamics of 292 genetically distinct strains. By integrating ResNet and StyleGAN2 to construct a Variational AutoEncoder (VAE) and utilizing a Siamese network for phenotypic similarity metrics, we efficiently encoded high-resolution microscopy images into 13-dimensional feature vectors, capturing phenotypic variability in aggregation patterns across time and strains. Human evaluation confirmed that our model’s reconstructions were visually indistinguishable from real images and closely aligned with input phenotypes. Importantly, the feature space is interpretable: individual dimensions correlate with biological features such as aggregate number and size, and extrapolation along these dimensions produces predictable morphological changes. Remarkably, our model revealed that developmental phenotypes are predictable from the earliest stages before visible aggregate formation begins. This predictability held across both genetic and environmental sources of variation, suggesting fundamental constraints on developmental trajectories and indicating that subtle phenotypic variations carry important information. These results demonstrate how machine learning can reveal hidden structure in complex multicellular dynamics and provide scalable methods for phenotypic analysis without manual annotation.

  PublisherOA PDFDOI

• POWER QUALITY ENHANCEMENT IN PV–PMSG GRIDS USING WHALE OPTIMIZATION-TUNED STATCOM

  International Journal of Apllied Mathematics · 2025-10-15 · 1 citations

  articleOpen access1st authorCorresponding

  The increasing penetration of renewable energy sources such as photovoltaic (PV) systems and permanent magnet synchronous generators (PMSGs) in distribution networks has introduced significant power quality challenges. This paper proposes a novel approach for improving the power quality of a grid-connected hybrid renewable system by employing a Static Synchronous Compensator (STATCOM) whose Proportional-Integral (PI) controller parameters are optimally tuned using the Whale Optimization Algorithm (WOA). Three distinct loading conditions — moderate load, heavy load surge, and light load — are considered to evaluate the system’s dynamic and steady-state performance. The optimized STATCOM effectively mitigates voltage dips, reduces power oscillations, and improves system transient response. Detailed time-domain simulations demonstrate that the proposed WOA-tuned STATCOM significantly enhances voltage stability, minimizes peak overshoot, improves settling time, and reduces harmonic distortions. Furthermore, the system performance under different loading scenarios confirms the superiority of the WOA-based optimization in stabilizing the hybrid renewable grid interface. The study also outlines directions for future expansion by considering nonlinear and unbalanced load impacts on system dynamics.

  PublisherOA PDFDOI

• Generative Modeling for Interpretable Anomaly Detection in Medical Imaging: Applications in Failure Detection and Data Curation

  Bioengineering · 2025-10-14 · 2 citations

  articleOpen access

  This work aims to leverage generative modeling-based anomaly detection to enhance interpretability in AI failure detection systems and to aid data curation for large repositories. For failure detection interpretability, this retrospective study utilized 3339 CT scans (525 patients), divided patient-wise into training, baseline test, and anomaly (having failure-causing attributes—e.g., needles, ascites) test datasets. For data curation, 112,120 ChestX-ray14 radiographs were used for training and 2036 radiographs from the Medical Imaging and Data Resource Center for testing, categorized as baseline or anomalous based on attribute alignment with ChestX-ray14. StyleGAN2 networks modeled the training distributions. Test images were reconstructed with backpropagation and scored using mean squared error (MSE) and Wasserstein distance (WD). Scores should be high for anomalous images, as StyleGAN2 cannot model unseen attributes. Area under the receiver operating characteristic curve (AUROC) evaluated anomaly detection, i.e., baseline and anomaly dataset differentiation. The proportion of highest-scoring patches containing needles or ascites assessed anomaly localization. Permutation tests determined statistical significance. StyleGAN2 did not reconstruct anomalous attributes (e.g., needles, ascites), enabling the unsupervised detection of these attributes: mean (±standard deviation) AUROCs were 0.86 (±0.13) for failure detection and 0.82 (±0.11) for data curation. 81% (±13%) of the needles and ascites were localized. WD outperformed MSE on CT (p &lt; 0.001), while MSE outperformed WD on radiography (p &lt; 0.001). Generative models detected anomalous image attributes, demonstrating promise for model failure detection interpretability and large-scale data curation.

  PublisherOA PDFDOI

• Molecular and cellular dynamics of measurable residual disease progression in myelodysplastic syndromes

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-10-19

  preprint

  Cancer relapse after treatment invariably proceeds from the persistence and progression of measurable residual disease (MRD), an ultrasmall malignant population. Despite the poor prognostic impact established by increasingly sensitive MRD detection methods, the molecular pathways defining MRD remain unknown. To identify the unique features of MRD and the molecular forces shaping its progression, we performed single-cell multi-omic profiling on DNA, RNA and protein layers for a longitudinal patient cohort with relapsed myelodysplastic syndromes (MDS) after stem cell transplantation (SCT), the only curative modality for MDS. We provide a comprehensive molecular portrait of MRD cells with novel markers, shared across genetically heterogeneous patients. MDS relapse after SCT manifested universally with marked phenotypic evolution. Genotype and phenotype analyses revealed MRD progression as a dynamic, evolutionary process rather than a static expansionary one, driven by both subclonal sweeping and cell state transitions. Malignant cells adapted to infiltrating T cells by rewiring IFN-γ responses to activate a key immunoevasive pathway. Our study demonstrates the power of longitudinal, single-cell multi-omic analysis for identifying, tracking, and understanding MRD cells, opening new avenues to target MRD persistence and progression.

  PublisherDOI

• Validation of portable in-clinic video-based gait analysis for prosthesis users

  Scientific Reports · 2024-02-15 · 22 citations

  articleOpen access

  Despite the common focus of gait in rehabilitation, there are few tools that allow quantitatively characterizing gait in the clinic. We recently described an algorithm, trained on a large dataset from our clinical gait analysis laboratory, which produces accurate cycle-by-cycle estimates of spatiotemporal gait parameters including step timing and walking velocity. Here, we demonstrate this system generalizes well to clinical care with a validation study on prosthetic users seen in therapy and outpatient clinics. Specifically, estimated walking velocity was similar to annotated 10-m walking velocities, and cadence and foot contact times closely mirrored our wearable sensor measurements. Additionally, we found that a 2D keypoint detector pretrained on largely able-bodied individuals struggles to localize prosthetic joints, particularly for those individuals with more proximal or bilateral amputations, but after training a prosthetic-specific joint detector video-based gait analysis also works on these individuals. Further work is required to validate the other outputs from our algorithm including sagittal plane joint angles and step length. Code for the gait transformer and the trained weights are available at https://github.com/peabody124/GaitTransformer .

  PublisherOA PDFDOI

Frequent coauthors

• Fabio Anselmi

  22 shared

• Weili Nie

  19 shared

• Richard G. Baraniuk

  16 shared

• Ryan Pyle

  Rice University

  14 shared

• McKell Woodland

  Rice University

  14 shared

• Yilong Ju

  Rice University

  13 shared

• Josue Ortega

  Yale University

  12 shared

• Andreas S. Tolias

  Baylor College of Medicine

  12 shared

Education

• Phd in Applied Mathematics

  Harvard University

  2008

• BA/SM, Applied Math / Computer Science

  Harvard University

  2002