We spent last weekend in Boston at the 2026 SNO/ASCO CNS Metastases Conference. The theme “From Moonshot to Groundshot: Transforming Science into Impact” turned out to be a fair description of the hallway conversations, not just the plenaries. Less “here is what might be possible,” more “here is what breaks when we try to do this on a Tuesday afternoon with a full reading list.” 

A lot of those conversations came back to our brain metastases model in NeuroQuant® Brain Tumor. Here’s what we took away. 

The treatment pathway got better, and harder, at the same time 

The good news dominated the science: CNS-active targeted agents and immunotherapies are meaningfully changing what happens after a brain mets diagnosis, and the radiation side of the field keeps refining its tools. 

The complication is that more good options mean more decisions. Several clinicians described the same dilemma in different words: when a patient presents with multiple lesions, the choice between SRS, systemic therapy with CNS penetration, surgery for a dominant symptomatic lesion, or some sequenced combination now depends on details that used to be secondary. How many lesions, exactly? How large is each one? Where are they, and how has each one moved since the last scan? 

Those are measurement questions. And measurement, it turns out, is where a lot of the frustration lives. 

Lesion counts are decision thresholds, and they’re fragile 

The count itself carries clinical weight — it can shape whether a patient is a candidate for focal treatment at all. But the count is not a stable number. It depends on scan protocol, slice thickness, contrast timing, reader experience, and how much time the reader has. A sub-centimeter lesion at the edge of detectability may be counted on one study and missed on the next, which then reads as new disease rather than a detection difference. 

We heard versions of this repeatedly at the booth: 

• Small lesion detection is the bottleneck. Confidence in whether a 3mm enhancing focus is real, and whether it was there last time, drives real decisions. 
• RANO-BM’s constraints are widely felt. The criteria were built for feasibility, but capping target lesions and relying on unidimensional measurement leaves a lot of information on the table — especially for patients with many small lesions, where a 1D measurement of the largest few doesn’t describe the disease burden well. 
• Longitudinal matching is manual and error-prone. Tracking which lesion is which across four or five timepoints, and doing it consistently, is tedious work that doesn’t scale with growing case volumes. 
• Volume changes precede diameter changes. Multiple clinicians made the point that volumetric trends give a more sensitive read on response, and that they’d act on them if they could get them reliably and quickly. 

Where NeuroQuant Brain Tumor fits 

Our brain metastases model performs automated segmentation and volumetric quantification of brain metastases, with results routed to PACS alongside the glioma and meningioma modules. In practice, the conversations that went deepest weren’t about the segmentation itself — they were about consistency. The value clinicians described was less “the AI found it” and more “the same method was applied to every timepoint, so when the volume moves, I believe the movement is real.” 

That framing matters. Automated volumetrics don’t replace a radiologist’s read; they reduce the variance underneath it, which is what makes a longitudinal trend interpretable. 

The other feature that drew consistent interest was RT STRUCT output. Given how much of the weekend’s discussion centered on stereotactic radiosurgery, this landed with a specific audience: when NeuroQuant Brain Tumor processes a study, it automatically generates contours for the segmentations as DICOM RT Structure Sets and returns them to PACS alongside the rest of the output. No separate request, no extra step — the contours are simply there when the study is. 

For patients with multiple targets, contouring is where a meaningful amount of manual time currently goes. It’s also a point where the diagnostic read and the treatment plan can quietly drift apart, since the lesions described in the report and the lesions contoured for planning are often delineated separately. Starting from a consistent, automatically generated set of contours narrows that gap. 

Where we’re going next 

The through-line from Boston was that the field is asking imaging to carry more decision weight than it used to, and the measurement infrastructure hasn’t fully caught up. That’s the gap we’re working on: better small-lesion sensitivity, dependable lesion-level tracking across timepoints, and outputs that fit the way tumor boards actually make decisions. 

Thanks to everyone who stopped by to talk, push back, and tell us what’s not working yet. Those conversations shape the roadmap more than anything else we do.