Varun Shijo

PhD Candidate - Computer Science and Engineering

340 Davis Hall, University at Buffalo, Buffalo, NY 14260-2500

varunshi@buffalo.eduORCIDScholarGitHub

Bio

Neural Representations for Real-time Computational Biosensing.

Varun Shijo's research compresses biosignal dimensionalities (1D EEG/IMU/speech, 2D/3D ultrasound and photoacoustic, and their pose/volume derivatives) into compact encodings that drive low-cost clinical sensing: reconstruction, detection, and real-time biofeedback. His framework spans acquisition and exploration, biomarker identification via learned features, and interpretation toward interventional methods, with translational clinical outcomes treated as the primary success metric. His current PhD work concentrates on two strands. Low-cost clinical sensing: adapting computer vision and edge inference to consumer-grade hardware for ultrasound tongue imaging (UltraSpeech, demoed at ASHA 2025), freehand 3D ultrasound reconstruction, and tumor detection in photoacoustic and ultrasound imaging. Multimodal biosignal encoding: learning representations across time-series and imagery modalities that retain interpretability and outperform rule-based baselines, moving toward closed-loop end-to-end systems for interventional use. An overarching goal drives the work: maximizing throughput between humans and computers via brain-computer interfaces.

Research

Neural Representations for Real-time Computational Biosensing.

Low-Cost Clinical Sensing

Computer vision and edge inference adapted to consumer-grade hardware for real-time biofeedback. UltraSpeech delivers ultrasound tongue imaging for speech therapy on CPU-only laptops. Freehand 3D ultrasound reconstruction targets self-directed breast cancer screening. Tumor detection and artifact reduction pipelines drive photoacoustic computed tomography toward translational deployment.

Multimodal Biosignal Encoding

Learned representations across time-series (EEG, IMU, speech) and imagery (ultrasound, photoacoustic) that retain interpretability and outperform rule-based baselines. Work spans MGNAT-SignalJEPA (NeurIPS 2025 EEG Foundation Challenge), photoacoustic reconstruction via Swin Transformers, and shared-tensor abstractions across sensing modalities. The target: closed-loop end-to-end systems suitable for interventional use.

Motivation

An overarching aim drives this work: maximizing throughput between humans and computers via brain-computer interfaces. Each modality (ultrasound, photoacoustics, EEG, IMUs, speech, tongue pose) contributes a slice of the bandwidth question, and the techniques (Transformers, CNNs, Mamba SSMs, multi-task pretraining) provide complementary models of thought.


6 works·47 citations·3 h-index·1 i10·updated 2026-05-12

Publications

Under Review

  • S. Hou, W. Bo, L. Cao, C. Liu, Varun Shijo, X. Zhang, L. Guo, Wenyao Xu (2026)

    Speech Annotation and Transcription Enhancer (SATE): An Automated System for Child Language Sample Analysis

    ACM IMWUT

    [submitted]

Journal Articles

Conference Proceedings

  • Varun Shijo, A. Das, W. Bo, S. Hou, L. Guo, Jun Xia, Wenyao Xu (2026)

    [1st]PULSE: A Principled Framework for Multimodal Speech Sound Disorder Assessment Using Ultrasound Tongue Imaging

    IEEE/ACM International Conference on Connected Health: Applications, Systems and Engineering Technologies (CHASE), Pittsburgh, PA (to appear in Elsevier Smart Health)

    [accepted]


Selected Projects

UltraSpeech

Real-time Ultrasound Tongue Imaging for Speech Therapy

Zero-to-one clinical biofeedback system built on a custom reactive DAG framework (sigflow-rt) that abstracts modalities by tensor shape and supports real-time inference inside the acquisition loop. The end-to-end pipeline captures B-mode ultrasound, audio, and video, runs ONNX inference for DeepLabCut tongue/lip pose, MediaPipe face mesh, and wav2vec2 phoneme transcription, renders live landmarks and a 3D tongue model, and scores tongue pose against UltraSuite TD references via Procrustes alignment. Zero-copy architecture (SIMD, shared memory, triple buffering) with async inference achieves real-time performance without a GPU.

  • 12–13 ms median latency on CPU only
  • 19,716 clinical inference operations validated
  • 14 tongue landmarks at 0.90 confidence
  • All streams synchronized via LSL for recording and playback
  • Demoed at ASHA Convention 2025 (Auspex Medix exhibitor)

sigflow-rt · ONNX · DeepLabCut · MediaPipe · wav2vec2 · LSL

MGNAT-SignalJEPA

NeurIPS 2025 EEG Foundation Challenge

Self-supervised EEG representation learning with a novel spatial block masking strategy. Built a comprehensive interpretability toolkit covering attribution analysis (Integrated Gradients, Grad-CAM), systematic ablation studies, and neuroscience-grounded validation.

  • Ranked 55/184 valid submissions
  • 1,202 total participants
  • Spatial block masking
  • Interpretability toolkit (Integrated Gradients, Grad-CAM, ablations)

PyTorch · EEG

Code


Honors & Awards

  • 3rd Place, Poster Presentation · RCBU Biomedical Ultrasound Symposium, Rochester, NY2025

Teaching

  • Spring 2025

    CSE574: Intro. to Machine Learning

    TA · Dr. Asif Imran

  • Fall 2024

    CSE573: Intro. to Computer Vision and Image Processing

    TA · Dr. Chen Wang

  • Summer 2024

    CSE573: Intro. to Computer Vision and Image Processing[instructor]

    Instructor of Record

  • Spring 2024

    CSE560: Data Models and Query Languages

    TA · Dr. Sreyasee Das Bhattacharjee

  • Spring 2023

    CSE666: Biometrics Image Analysis

    TA · Dr. Nalini Ratha


Mentoring

Graduate

  • Nihar Asare(MS, Robotics @ UB)

    Wireless self-directed 3D freehand breast ultrasound

  • Arianna Dougherty(MS, BE @ UB)

    Breast PACT for tumor detection and localization

Undergraduate

  • Emma Durham(Senior UG, BE @ UB)

    Robotic-arm position tracking for self-directed ultrasound scanning

  • Hannah Pham(Senior UG, BE @ UB)

    Face spoofing detection using SWIR imaging

  • Chaeyeon Kim(UG, IE → CSE @ UB)

    BCI fundamentals, OpenBCI Ganglion recording

    Team placed 2nd at UBHacking 2025 (SSVEP / motor-imagery game UI)

  • Nhat Dinh(UG, CompEng @ UB)

    UltraSpeech: finetuning experiments and PyQt acquisition app

High School

  • Michelle Lin(HS, Williamsville North)

    Camera-based PPG for pulsatile signal estimation

  • Emma Zhang(HS, Williamsville North)

    SWIR for moisture quantification


Service

Journal Reviewer

  • ACM Transactions on Computing for Healthcare (11 manuscripts)2022–2025
  • PLOS ONE2025
  • Elsevier Smart Health2024
  • World Scientific JIOHS2023

Conference Reviewer

  • IEEE EMBS Conference on Biomedical and Health Informatics (BHI)2024, 2025
  • IEEE/ACM Conference on Connected Health (CHASE)2025
  • IEEE EMBS Body Sensor Networks (BSN)2024

Technical Program Committee

  • IEEE EMBS Body Sensor Networks (BSN)2023

Education

  • PhD in Computer Science and Engineering

    University at Buffalo 2022 - Present

    Admitted to Candidacy: September 29, 2025

  • MS in Computer Science - Artificial Intelligence

    University at Buffalo 2019

  • BEng in Information Technology

    University of Mumbai 2017