Differentiable Solutions provides an embedded C++ SDK for medical-imaging OEMs to integrate AI-powered CBCT segmentation directly into their own clinical software. The SDK runs entirely in-process on a workstation GPU, with no cloud dependency, and currently supports over 70 classes of dental anatomy in a single inference pass. It includes a regulatory dossier package to streamline CE MDR or FDA 510(k) submissions.
Funding
Funding not disclosed
Founders
Product
Problem
Medical-imaging OEMs face multi-year, high-cost efforts to build in-house AI segmentation capabilities, requiring specialist teams in C++, CUDA, 3D imaging, MLOps, and regulatory affairs. Even after development, achieving a dossier-ready binary with validated reproducibility and cross-platform parity adds another 24 to 30 months before clinical deployment.
Solution
Differentiable Solutions licenses a production-grade C++ binary SDK that OEMs embed directly into their own viewers and planning suites, running in-process on the workstation GPU with no cloud, container, or service dependency. The SDK performs semantic segmentation of dental cone-beam CT volumes across more than 70 anatomical classes in a single inference pass, with a full clinical CBCT processed in under a minute on a modest RTX A4000 GPU. The company delivers the segmentation engine along with a component-level regulatory dossier—including SOUP, model card, validation summary, and IFU template—that plugs directly into the OEM's CE MDR or FDA 510(k) technical file. The SDK is built from first-principles applied mathematics, with encrypted, licence-bound model artifacts and every release traceable to a sealed, quality-gated training run.
Target Audience
Primary customers are medical-imaging OEMs building clinical viewers, planning suites, or diagnostic products who need embedded segmentation capabilities without building the AI and regulatory infrastructure themselves. The company also serves clinical partners and KOLs through validation studies and dataset licensing on retrospective anonymised data.
Features
- In-process binary C++ SDK for Windows and Linux with fully encapsulated public API and no third-party headers in the customer's build
- Single-pass semantic segmentation of more than 70 classes including teeth at FDI granularity, individual pulps, jawbone, mandibular canals, maxillary sinuses, pharynx, and dental work
- GPU-accelerated inference on a single modest workstation card (RTX A4000) completing a full clinical CBCT in under a minute, with a memory-saving mode for GPUs with less than 12 GB
- Regulatory dossier package (SOUP, model card, validation summary, IFU template) designed to plug into CE MDR or FDA 510(k) technical files
- Model IP protection through encrypted, licence-bound artifacts, software bill of materials per release, and security posture assessed against IEC 62443, ISO/IEC 27001, and the EU Cyber Resilience Act
- Reproducible releases with every model uniquely versioned and traceable to a sealed, quality-gated training run
- Optional gRPC/REST service deployment with reference clients in Python, C# and Java for non-in-process integration needs