Shahariar Rabby

Rabby, Chaudhary, Saha, Nakhmani, Zhang, Barr, Reinhardt, Bhatt, Bodduluri. Light Convolutional Neural Network for Detecting COPD: A Multicenter Model Development and External Validation

medRxiv, 2025 · under review at The Lancet Digital Health · First & corresponding author

A lightweight 3D CNN (10.6M parameters) for population-scale COPD detection from inspiratory chest CT. Trained on COPDGene (21 centers, 13K+ CT scans); AUC 0.92 internally, with external validation on SPIROMICS (AUC 0.92) and NLST (AUC 0.82).

@article{rabby2025copd,
  title   = {Light Convolutional Neural Network for Detecting Chronic Obstructive Pulmonary Disease: A Multicenter Model Development and External Validation},
  author  = {Rabby, Shahariar and Chaudhary and Saha and Nakhmani and Zhang, Chengcui and Barr and Reinhardt and Bhatt and Bodduluri, Sandeep},
  journal = {medRxiv},
  year    = {2025},
  doi     = {10.1101/2025.07.30.25332459}
}

Rabby, Zhang, Nakhmani, Bodduluri, Bhatt. STRIPE: Directional Feature Extraction for Pulmonary Fissure Segmentation In preparation

Manuscript in preparation · First author

A directional encoder–decoder (AFM + DDA; LCNet-100 encoder; UNet++-like decoder) for pulmonary fissure segmentation. Systematic ablation shows encoder-level directional modeling preserves thin-structure continuity: Dice 0.873, HD95 0.72 mm, 5.15M parameters, evaluated on 50 COPDGene CT scans (5-fold CV).

Saha, Chaudhary, Rabby, Sthanam, Bodduluri, Nakhmani, Zhang, Bhatt. Machine Learning Based Detection of True Ventilatory Restriction

Respiratory Medicine (Elsevier), 2025

Rabby, Saha, Nakhmani, Reinhardt, Zhang, Bhatt, Bodduluri. COPDxNet: An End-to-End Deep Neural Network for COPD Detection

American Thoracic Society (ATS) International Conference, 2024 (abstract)

Rabby, Zhang. THREAD: Efficient Topology-Preserving Network for Thin Structure Segmentation In preparation

Manuscript in preparation · First author

Topology-preserving architecture with a Directional Attention Bridge, Sobel-initialized Topology-Preserving Gate, and Cascaded Directional Refinement. On Crack500: F1 0.7940, clDice 0.8231, 48.6% fewer FLOPs; 92 FPS on GPU and 29.6 FPS on a Raspberry Pi 5 (~19× faster than SSMs). Cross-domain transfer: Dice 0.866 on pulmonary fissures without retraining.

Rabby, Zhang. dCrack: Enhancing Fine-Grained Crack Segmentation with Edge-Guided Attention ✦ dCrack61K

IEEE International Conference on Information Reuse and Integration (IRI), 2025 · First author

Edge-Guided Attention for minority-class crack segmentation, plus the dCrack61K benchmark (61,943 images, 36 sources, 2,000+ corrected annotations). Cross-domain transfer: F1 0.8251 on CHASE_DB1 retinal vessels without vessel-specific training.

@inproceedings{rabby2025dcrack,
  title     = {dCrack: Enhancing Fine-Grained Crack Segmentation with Edge-Guided Attention},
  author    = {Rabby, Shahariar and Zhang, Chengcui},
  booktitle = {IEEE International Conference on Information Reuse and Integration (IRI)},
  year      = {2025}
}

Islam, Hasan, Ovi, Rabby, Rahman. BornoDrishti: Leveraging Vision Encoders and Domain-Adaptive Learning for Bangla OCR on Diverse Documents Published

EACL Industry Track, pp. 278–286, 2026

Hasan, Islam, Rabby, Rahman. Beyond Characters: Position-Aware Metric and a Lightweight LLM for Low-Resource Bangla Punctuation Restoration

IEEE International Conference on Big Data, 2025

Rabby, Ali, Islam, Abujar, Rahman. Enhancement of Bengali OCR by Specialized Models and Advanced Techniques for Diverse Document Types

IEEE/CVF WACV Workshops, 2024 · First author

Akanda, Ahmed, Rabby, Rahman. Optimum Deep Learning Method for Document Layout Analysis in Low Resource Languages

ACM Southeast Conference, 2024

Ali, Rabby, Islam, Mahamud, Hasan, Rahman. Gold Standard Bangla OCR Dataset: An In-Depth Look at Data Preprocessing and Annotation Processes

EMNLP Industry Track, 2023 · Co-first author

Zhao, Zhang, Saxena, Wallach, Rabby. Ballot Tabulation Using Deep Learning

IEEE IRI, 2023 · NSF #2154589

Rabby, Haque, Islam, Abujar, Hossain. Ekush: A Multipurpose & Multitype Comprehensive Database for Bangla Handwritten Characters ✦ Best Paper

RTIP2R, Springer, 2018 · First author

A large-scale Bangla handwritten character dataset: 673K instances across 242 classes. A common baseline dataset for Bangla OCR research.

Islam, Rahman, Rabby. Sankhya: An Unbiased Benchmark for Bangla Handwritten Digits Recognition

IEEE Big Data, 2019

Islam, Khushbu, Rabby, Hossain. Ishara-Lipi: The First Complete Multipurpose Dataset for Bangla Sign Language

ICBSLP, IEEE, 2018