About NucliVision

Built inside nuclear medicine departments. Designed to change how radiologists read scans.

Where this started

In 2021, Maarten Larmuseau was interpreting a Lu-177 DOTATATE therapy scan at Massachusetts General Hospital. The patient had metastatic neuroendocrine tumor. The scan was technically adequate - but the image noise made three small hepatic lesions ambiguous. Were they responding to therapy? The answer mattered for dosing the next cycle.

He pulled up the prior scan. Different scanner protocol. Different reconstruction parameters. No easy comparison. He spent 40 minutes with image post-processing tools that were designed for diagnostic CT, not SPECT.

Three months later, he brought that case to a room full of engineers at MIT. NucliVision started that afternoon.

NucliVision origin

What we are building toward

Nuclear medicine is the only imaging modality that measures molecular function - not just anatomy. A PET scan can detect tumor metabolism months before a mass appears on CT. SPECT dosimetry can personalize radiopharmaceutical therapy doses the way oncology titrates chemotherapy. The clinical data is there. The tools to use it efficiently are not.

NucliVision writes software that fills that gap. Not general-purpose medical AI. Specifically: AI trained on nuclear medicine physics, nuclear medicine acquisition protocols, and the specific reporting needs of nuclear medicine physicians. Nuclivision is backed by LUMO Labs through a seed round investment.

How we work

Physics-grounded AI

Every model we build starts with nuclear medicine physics constraints. Scatter correction, attenuation maps, dead-time correction - these aren't artifacts to smooth over, they're inputs to the algorithm.

Clinician-first design

Our product team includes nuclear medicine physicians who read scans weekly. Features are tested in real reading rooms before they reach a release candidate. No hospital buys a product a physician won't use.

Regulatory clarity

We treat FDA 510(k) clearance as a product milestone, not a legal obstacle. Clinical validation data is collected prospectively. IEC 62304 documentation is current. We don't launch without regulatory approval.

Company milestones

2021

Founded

NucliVision incorporated in Boston, MA. Founding team: two nuclear medicine physicians and three ML engineers from MIT CSAIL.

2022

First clinical dataset

Data access agreement signed with two academic medical centers. 120,000 de-identified nuclear medicine studies secured for model training and validation.

2023

Alpha clinical testing

Image enhancement model deployed in parallel-read study at Beth Israel Deaconess Medical Center. Blinded reader study completed with 8 nuclear medicine physicians.

2024

Platform v1.0 and Seed Round

Full platform including dosimetry calculation and report drafting released. Seed Round closed. FDA pre-submission meeting completed.