# Torch Recon: An Innovative Reconstruction Software for Increased Throughput and Improved Low-Count Quantitative SPECT Imaging

> **NIH NIH R44** · VOXIMETRY, INC. · 2024 · $1,003,632

## Abstract

PROJECT SUMMARY/ABSTRACT
Single-photon emission computed tomography (SPECT) imaging plays a pivotal role in radiopharmaceutical
therapy (RPT), allowing clinicians to personalize prescriptions and assess treatment response. However,
traditional SPECT reconstruction methods often encounter challenges related to noise, artifacts, and lengthy
processing times. In imaging, accurate correction of scattering effects and enhancement of signal-to-noise ratio
(SNR) are critical for achieving quantitively accurate images required for dosimetry guidance of RPT and
treatment response assessment. Torch Recon is a cutting-edge software that harnesses the synergistic power
of Monte Carlo simulation and deep learning techniques to address the limitations of conventional reconstruction
methods. Monte Carlo simulation accurately models photon interactions within tissues, leading to improved
accuracy and resolution in the reconstructed images. Complementing this, deep learning algorithms are
employed to enhance image quality, reduce noise, and suppress artifacts. These algorithms leverage large
datasets to learn intricate patterns and relationships, resulting in sharper, more informative SPECT images.
Torch Recon represents the fusion of Monte Carlo simulation and deep learning, enabling a dynamic and
adaptive reconstruction process which has the potential to not only improve quantitative SPECT but also SNR
which is especially important for scenarios with low counting statistics, e.g., alpha emitters. As part of a previous
Phase I contract, we incorporated a SPECT reconstruction algorithm with a GPU-accelerated Monte Carlo-based
scatter estimator into the GPU-based Torch software system for RPT dosimetry. In this Phase II proposal, we
will (1) implement AI denoising techniques into Torch Recon, (2) assess accuracy and performance of the
reconstruction software using phantoms, and (3) validate clinical usability and effectiveness through a
prospective clinical trial. By completing the milestones of this Phase II proposal, Torch Recon will be ready for
510(k) clearance and commercialization.

## Key facts

- **NIH application ID:** 10922106
- **Project number:** 1R44CA291564-01
- **Recipient organization:** VOXIMETRY, INC.
- **Principal Investigator:** Joseph Grudzinski
- **Activity code:** R44 (R01, R21, SBIR, etc.)
- **Funding institute:** NIH
- **Fiscal year:** 2024
- **Award amount:** $1,003,632
- **Award type:** 1
- **Project period:** 2024-09-19 → 2026-08-31

## Primary source

NIH RePORTER: https://reporter.nih.gov/project-details/10922106

## Citation

> US National Institutes of Health, RePORTER application 10922106, Torch Recon: An Innovative Reconstruction Software for Increased Throughput and Improved Low-Count Quantitative SPECT Imaging (1R44CA291564-01). Retrieved via AI Analytics 2026-05-23 from https://api.ai-analytics.org/grant/nih/10922106. Licensed CC0.

---

*[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*