Publications | Saurav Dosi

2026

arXiv

IntelX: Autonomous Indoor Exploration via Local Coverage Gain Prediction and Structure-Aware Planning

Saurav Dosi, Itay Kadosh, and Yu Xiang

2026

Under review

Abstract Video Code Documentation Live Demo Website

IntelX is an autonomous indoor exploration framework that, instead of completing the unseen map with a generative model, predicts how much new area each nearby goal would reveal. A single lightweight U-Net regresses this coverage gain densely over a 20 m robot-frame occupancy window in 8.1 ms, supervised by a simulation oracle that measures the coverage each reachable goal actually yields. Junctions show up in the same field as multi-peaked angular structure, which a global planner turns into branches that it clears weakest-first. On three held-out floor plans and three out-of-distribution corpora, IntelX reaches 99 percent coverage on large maps with paths at least 20 percent shorter than every baseline that also gets there.
Med Phys
Inter-Crystal Scatter Recovery of a Light-Sharing Depth-Encoding Prism-PET Prototype Scanner Using a Diffusion Model

Wanbin Tan, Saurav Dosi, Soroush Shabani Sichani, Yixin Li, Zipai Wang, Eric Petersen, Xinjie Zeng, Jinyi Qi, and Amir H. Goldan

Medical Physics, 53(9), e70661, 2026

Abstract DOI PubMed BibTeX Video Website

Inter-crystal scattering (ICS) is a fundamental limitation in ultra-high resolution PET, particularly in light-sharing detectors with small crystal elements, where Compton scatter across neighboring crystals obscures the initial interaction position and degrades spatial resolution, contrast, and quantitative accuracy. This work develops and experimentally validates a scanner-level ICS recovery framework for a light-sharing depth-encoding Prism-PET prototype scanner. We build a training dataset by combining scanner-level simulations with experimental measurements, and propose D3Recovery, a diffusion-driven, physics-informed tripartite framework that estimates the 3D position and deposited energy of the initial interaction from 8x8 SiPM signal patterns. D3Recovery reaches 70.1 percent crystal identification accuracy for ICS events, improves DOI estimation to 3.6 mm FWHM, and halves mean crystal localization error versus center-of-gravity positioning (2.18 mm to 0.99 mm). On Derenzo and Hoffman brain phantoms it improves peak-to-valley ratio, MAE, SSIM and noise, retaining roughly 70 percent of otherwise discarded coincidences for a better resolution-sensitivity balance in low-dose imaging.
@article{Tan2026D3Recovery, author = {Tan, Wanbin and Dosi, Saurav and Shabani Sichani, Soroush and Li, Yixin and Wang, Zipai and Petersen, Eric and Zeng, Xinjie and Qi, Jinyi and Goldan, Amir H.}, title = {Inter-Crystal Scatter Recovery of a Light-Sharing Depth-Encoding Prism-PET Prototype Scanner Using a Diffusion Model}, journal = {Medical Physics}, volume = {53}, number = {9}, pages = {e70661}, year = {2026}, month = sep, publisher = {Wiley}, doi = {10.1002/mp.70661}, pmid = {42740403}, }

SNMMI

D3RecoveryMux: Inter-Crystal Scatter Recovery for TOF-DOI Prism-PET Detectors with Multiplexed Readout Using Diffusion Model

Saurav Dosi, Wanbin Tan, Soroush Shabani Sichani, Yixin Li, Zipai Wang, Eric Petersen, and Amir H. Goldan

Spotlight · Journal of Nuclear Medicine, 67(Supplement 1), 261694, 2026

SNMMI 2026 Annual Meeting

Abstract BibTeX Article

@article{Dosi2026D3RecoveryMux,
  author = {Dosi, Saurav and Tan, Wanbin and Shabani Sichani, Soroush and Li, Yixin and Wang, Zipai and Petersen, Eric and Goldan, Amir H.},
  title = {D3RecoveryMux: Inter-Crystal Scatter Recovery for TOF-DOI Prism-PET Detectors with Multiplexed Readout Using Diffusion Model},
  journal = {Journal of Nuclear Medicine},
  volume = {67},
  number = {Supplement 1},
  pages = {261694},
  year = {2026},
  month = may,
  note = {SNMMI 2026 Annual Meeting},
}

SNMMI

D3Recovery: Inter-Crystal Scatter Recovery for TOF-DOI Prism-PET Scanner Using Diffusion Model

Wanbin Tan, Saurav Dosi, Soroush Shabani Sichani, Yixin Li, Zipai Wang, Eric Petersen, and Amir H. Goldan

Oral Presentation · Journal of Nuclear Medicine, 67(Supplement 1), 261669, 2026

SNMMI 2026 Annual Meeting

Abstract BibTeX Article Video Website

@article{Tan2026D3RecoverySNMMI,
  author = {Tan, Wanbin and Dosi, Saurav and Shabani Sichani, Soroush and Li, Yixin and Wang, Zipai and Petersen, Eric and Goldan, Amir H.},
  title = {D3Recovery: Inter-Crystal Scatter Recovery for TOF-DOI Prism-PET Scanner Using Diffusion Model},
  journal = {Journal of Nuclear Medicine},
  volume = {67},
  number = {Supplement 1},
  pages = {261669},
  year = {2026},
  month = may,
  note = {SNMMI 2026 Annual Meeting},
}

2022

Springer
Segregation of Areca Nuts Using Three Band Photometry and Deep Neural Network

Saurav Dosi^*, Bala Vamsi^*, Samarth S. Raut, and D. Narasimha

In Soft Computing and its Engineering Applications, Springer, 2022

Abstract DOI BibTeX PDF

In the Malnadu region of Karnataka state in India, often tender Areca nuts of five to eight months old are harvested together, manually separated according to ripeness based on their colour, dehusked, sliced and boiled before marketing. This is a tedious process involving skilled labour and due to the remnant pesticides on the arecanut surface, the health of the worker is adversely affected. Hence to aid the farmers, a technique is developed to carry the task of segregation based on Three Band Photometry and Machine Learning. Light green to fully ripe orange Areca nuts from two different regions were used for Training as well as Testing, using state-of-the-art YOLOv3 Object Detection Deep Neural Network model. They were studied at various stages after plucking between fresh from harvest to one week old. The method developed was found to be effective for both samples at any of these stages.
@inproceedings{10.1007/978-3-031-05767-0_2, doi = {10.1007/978-3-031-05767-0_2}, author = {Dosi, Saurav and Vamsi, Bala and Raut, Samarth S. and Narasimha, D.}, editor = {Patel, Kanubhai K. and Doctor, Gayatri and Patel, Atul and Lingras, Pawan}, title = {Segregation of Areca Nuts Using Three Band Photometry and Deep Neural Network}, booktitle = {Soft Computing and its Engineering Applications, Springer}, year = {2022}, publisher = {Springer International Publishing}, address = {Cham}, pages = {15--27}, isbn = {978-3-031-05767-0}, }