Lunit

Lunit is a medical imaging AI company run by a Seoul-listed Korean corporation. Its main business is discriminative models — flagging suspicious lesions and producing malignancy scores on mammograms and chest X-rays — rather than a chatbot that writes text. Customers are imaging centers at mid-size and large hospitals, regional screening programs, and oncology pharma companies; individual users do not even have a sign-up entry point.

Core capabilities

• Mammography reading: overlays lesion heatmaps and a 0–100 malignancy likelihood score inside PACS, outputs BI-RADS density grades, and auto-tags normal studies so they can skip reading
• Chest X-ray triage: dual engines for abnormality detection and normal determination flag more than ten categories of abnormality — pleural effusion, pneumothorax, nodules and others — in a single pass, with prior-versus-current comparison
• 3D tomosynthesis: automatically picks out the key slices, cutting reading fatigue under double-read workflows
• Computational pathology: the SCOPE line performs spatial tumor-microenvironment analysis and HER2 and PD-L1 quantification on H&E and IHC slides, sold to pharma companies and CROs
• Generative side lines: a VLM drafts chest X-ray reports for physicians to sign off, and a medical-science foundation model with 16 billion parameters provides clinical decision support

Where it fits Breast cancer screening programs in Europe and Australia that use double reading are the best fit: AI takes the place of the second reader and immediately halves the human workload. Next come radiology departments and imaging centers with high chest X-ray throughput, and oncology pharma companies doing companion diagnostics; organizations without PACS and individual self-serve users are out of reach.

Unique advantages Competitors' single-point AUCs sit within the same order of magnitude; Lunit's differentiation is not accuracy but three things: prospective population-level evidence, the end-to-end screening workflow filled in after the Volpara acquisition, and distribution bundled with equipment from GE, Philips, Hologic and others.

Editorial assessment The evidence is top-tier: ScreenTrustCAD prospectively enrolled 55,851 people, and one physician plus AI was non-inferior to two physicians; but the study was partly vendor-funded and conducted at a single Swedish hospital. What most deserves remembering is the gap between clearances and marketing — all six of the company's FDA 510(k) clearances went the substantial-equivalence route, with not a single De Novo, and in the US the chest X-ray product is actually cleared only for triage notification of two findings, pleural effusion and pneumothorax, while the website advertises 11 categories of abnormality. The finances are also tight: in 2025 revenue and operating loss were both KRW 83.1 billion, and the current ratio was about 0.26. There is a wholly owned subsidiary in Shanghai, but no Chinese-language site, no ICP filing and no NMPA registration certificate, so it has not yet entered Chinese clinical practice.