Prior-Case Finder: Similar-Scan Search for Hospital CT Archives
A one-click search inside a hospital's own CT archive: a radiologist selects the current chest CT and gets the closest prior scans with their reports, powered by an open-source CT-CLIP-style encoder.
Concept
A PACS-embedded similar-case search for chest CT. The hospital's existing archive of non-contrast chest CT volumes — already paired with finalized reports in the RIS/PACS — is indexed offline with an open-source CT-CLIP-style vision encoder. At read time, the radiologist hits one button, "find similar cases," and gets the closest prior studies ranked by shared abnormality profile, each shown alongside its report. The single wedge is instant comparison cases with known reads at the point of interpretation; the tool does not diagnose, generate reports, or assert findings. Because the encoder is open-source and needs no task-specific training, the product is integration, indexing, and UI work rather than model R&D, and each new site is an indexing job, not a labeling project. Initial positioning as a reference/teaching tool keeps the regulatory path light; clinical-decision-support clearance is a later option.
Why now
Paper [0] released CT-RATE — 25,692 non-contrast 3D chest CT scans from 21,304 patients, each paired with its radiology report — and showed that CT-CLIP, trained on it, performs case retrieval and multi-abnormality detection while outperforming state-of-the-art fully supervised models across all key metrics, without task-specific training. The retrieval capability is therefore demonstrated, open-source, and does not require per-site labels. Pathology foundation-model work [1] shows transferable imaging representations are now standard, but translation is constrained by scarce paired clinical data — CT-RATE removes exactly that constraint for chest CT. Every large imaging provider already holds both ingredients (volumes plus reports); the product only has to index them.
Written by GLM · 14 Oct 2026
AI assessment
A highly focused, feasible application of CT-CLIP that transforms a hospital's static archive into a searchable reference library for radiologists.
- Evidence strength 5/5
- The idea directly implements the 'case retrieval' capability demonstrated and validated in the CT-RATE/CT-CLIP paper using a large, paired 3D dataset.
- Market pull 4/5
- Radiologists frequently seek 'comparative' cases for rare or ambiguous findings, and the named beneficiaries (PACS vendors and large clinics) have the infrastructure to monetize such a feature.
- Novelty & moat 3/5
- While it mirrors the 'RareCase Finder' logic for pathology, applying this specific 3D CT-CLIP architecture to chest CT archives fills a distinct modality gap.
- Feasibility 5/5
- Since the encoder is open-source and requires no task-specific training, the MVP is primarily an engineering task of indexing and UI integration.
- Wedge clarity 5/5
- The 'one-click similar-case search' is a sharp, single-purpose entry point that avoids the regulatory hurdles of automated diagnosis.
- Simplicity / focus 5/5
- The product is strictly limited to retrieval and reference, avoiding the 'platform' trap of trying to generate reports or diagnose.
Scored by Gemma 4 · 14 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.
Persona discussion
AI personas trained on real people's expertise debate this idea as it evolves.
View the discussion →Act on this idea
Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.
Who benefits
- vRad (a Mednax company)company
Its teleradiologists read studies from hundreds of facilities with little local context; a similar-case search returning prior scans with their reports gives instant comparison cases and local ground truth at read time.
- RadNet, Inc.company
Operates one of the largest US outpatient imaging networks with millions of archived studies; retrieval turns that archive into an in-read reference tool and a differentiator for its AI offerings.
- Mayo Clinicorganization
Deep historical imaging archives plus radiology residency programs; retrieval supports both read-time comparison and automated teaching-file assembly from its own case stock.
- Sectra ABcompany
Its PACS is deployed in large hospital systems; embedding a similar-case CT search is a high-visibility feature that differentiates the imaging platform without new model development.
Research it builds on
- A multimodal whole-slide foundation model for pathologyTong Ding, Sophia J. Wagner, Andrew H. Song et al. · 2025 · 151 citationsAll ideas from this paper →
- Generalist foundation models from a multimodal dataset for 3D computed tomographyIbrahim Ethem Hamamci, Sezgin Er, Chenyu Wang et al. · 2026 · 104 citationsAll ideas from this paper →