MEDICAL IP

The segmentation models are the engine of the core modules, and the portfolio carries value beyond them. DeepCatch (body composition from CT), the DeepCatch X measurement modules on chest X-ray, DeepFore (liver cancer risk prediction) and TiSepX (lung volumetry) are the models themselves.

The flagship MEDIP PRO is a medical image processing workstation: its 510(k) summary (K191026, 2019) describes 3D reconstruction and visualization, automatic segmentation for specific targets alongside user directed segmentation and measurement tools, and export for 3D printing, with a conventional segmentation product (Materialise's Mimics inPrint) as its predicate.

The company also sells a 3D printing service for anatomical models (ANATDEL) and anatomy education products, including virtual reality content, whose value to a medical school does not depend on a model running at the point of sale.

A clinician directs and accepts the output of the segmentation products, and the prediction product is the open question. MEDIP PRO's 510(k) summary (K191026) describes automatic segmentation for specific targets combined with user directed segmentation and measurement tools, for pre operative planning, and states that its 3D printed models are for non diagnostic use; the clinician edits and accepts what the software produces.

DeepFore, approved in Korea at Class III, predicts whether liver cancer will occur and, in DeepFore Recur, whether it will recur; the home page gives its predictive accuracy as 89 percent. Prediction against a future event is a different oversight problem from measurement, and no threshold, handling of borderline predictions or rule for what a clinician does with a low risk result is published.

The method is public for one product and general for the rest. DeepFore's model, which predicts hepatocellular carcinoma from CT based imaging biomarkers in patients with chronic hepatitis B, is described in a peer reviewed paper in the Journal of Hepatology (2025), where the model and its validation are published.

MEDIP PRO's 510(k) summary (K191026) describes automatic segmentation for specific targets, image quality improvement and texture analysis without naming a model type, and no architecture, training data description or versioning practice is published for the segmentation and measurement products. The company's publications page lists its research.

The model layer is short. MEDICAL IP builds its own segmentation, quantification and prediction models, cleared or approved as medical devices in the United States (MEDIP PRO, K191026; DeepCatch, K223556), Korea and elsewhere, so a model cannot be swapped for another supplier's without going back through those regulators. MEDIP PRO is described in its 510(k) summary as stand alone software.

The company publicizes an NVIDIA Omniverse connection and a relationship with AWS, and names no hosting provider, region or subprocessor for any cloud component. Anatomical models printed through its ANATDEL service leave the institution as physical objects, and who manufactures them and how the source files are handled is not stated.

The published record is real and uneven across the portfolio. DeepFore's CT based model for predicting hepatocellular carcinoma in patients with chronic hepatitis B was published in the Journal of Hepatology in 2025 and featured on the cover of the June issue; the home page gives its predictive accuracy as 89 percent.

For the flagship MEDIP PRO, a retrospective study in PLOS ONE (2026) used its semi automatic segmentation of orbital structures on CT to assess thyroid eye disease activity and reported an AUC of 0.938 when the measurements were combined with clinical parameters, and the 510(k) summary (K191026) states that no clinical studies were performed for clearance.

A randomized study compared virtual anatomy dissection with cadaver dissection for the education products, and clinical results with patient specific 3D kidney models in robotic partial nephrectomy have been published. MEDICAL IP is a Seoul National University Hospital spin off, and much of the clinical work comes from that institution.

Nothing on the product pages or in MEDIP PRO's 510(k) summary states how patient data is retained, de identified or used for model training, and the site footer links a privacy policy and terms of service, the published documents where a position on clinical images would sit. MEDIP PRO runs as stand alone software, so for that product the studies stay where it is installed. Two stewardship questions are specific to this portfolio.

Anatomical models printed from one person's imaging are identifiable health information in physical form that leaves the institution and cannot be encrypted, logged or remotely deleted, and nothing published addresses who makes them, how they are shipped or how the source file is handled afterward. And if the anatomy education content derives from real patient imaging rather than consented or synthetic reference anatomy, that is secondary use in a commercial product. Safety engineering, meaning guardrails and how a safety event is handled, is not published.

No business associate status is stated on the product pages or in the 510(k) summaries, and the site footer links a privacy policy and terms of service, which are where the company's role for the service would be set out. The company holds two US clearances, has a US contact office and names US academic users (Mayo Clinic, UCLA, Arizona State University), so US obligations attach to those deployments.

It also operates under Korean and Chinese regimes, with a Chinese certification and deployment in a Chinese hospital, and Chinese law is the most restrictive of the three on moving health data across borders. MEDIP PRO runs as stand alone software, which narrows what the company receives for that product; cloud components and the 3D printing service are where studies would leave the hospital.

The certification MEDICAL IP names is ISO 13485, obtained with its European CE mark in 2019: a quality management standard for medical device manufacturers, not an information security audit. No ISO 27001 certificate, SOC 2 report, trust center, security page or penetration test summary is published, and the site footer links only a privacy policy and terms of service, the remaining published documents where a Korean information security certification or an outside check would be named.

MEDICAL IP's US position is identifiable in the FDA database. MEDIP PRO was cleared on 7 November 2019 (K191026, 21 CFR 892.2050, product code LLZ) as software that transfers DICOM imaging from a scanner to an output file with image segmentation, also for pre operative planning, with 3D printed models for non diagnostic use; the predicate was Materialise's Mimics inPrint. DeepCatch was cleared on 16 June 2023 (K223556, 892.2050, product code QIH).

Outside the US the company reports European CE marking under the Medical Device Directive with ISO 13485 (December 2019), UK MHRA registration (September 2021), a Chinese NMPA certificate for MEDIP PRO (April 2025) and Korean MFDS approvals from Class I (ANATDEL) to Class III (MEDIP Pneumovid, DeepFore and DeepFore Recur). DeepFore, DeepCatch X and TiSepX are not among the US clearances.

The company writes FDA certification; the FDA clears Class II devices through 510(k) and does not certify them. The European mark was issued under the directive since replaced by the EU Medical Device Regulation, and the company gives no current MDR status.

No results by patient group are published for any product. MEDIP PRO's 510(k) summary reports phantom and bench testing with no subgroup analysis. The gap has a specific shape here: segmentation, body composition measurement and organ volumetry depend on anatomy that varies across populations, and reference ranges derived from one population do not transfer automatically to another.

The company's clinical research comes largely from Korean institutions, and its products are sold in the United States, China, the United Kingdom and Europe, so a normal range that is slightly wrong for a population produces measurements that are systematically wrong rather than visibly wrong, most of all in the DeepCatch X measurement modules. No governance framework or monitoring commitment is published.

Mechanisms exist that let a clinician check an output, and nothing stands behind it. MEDIP PRO combines automatic and user directed segmentation, and the clinician edits and accepts the result; its 510(k) summary (K191026) reports phantom distance accuracy and segmentation and usability testing against the predicate with no accuracy figures, and states that no clinical studies were performed, so it gives no error rate a buyer can hold the product to.

The DeepCatch summary (K223556, 2023) is the other US filing in the segmentation and quantification family. No warranty, indemnity or remediation commitment is published. Segmentation accuracy feeds everything downstream in this portfolio, including printed models, so an error there can end up in a physical object nobody re measures.

MEDICAL IP works on standard imaging and names its users. MEDIP PRO's 510(k) summary describes it as software for transferring DICOM imaging from a scanner, and the home page lists DICOM with OBJ, STL, 3MF and VTK formats; named users include Mayo Clinic, UCLA, Arizona State University, Seoul National University Hospital, Samsung Medical Center, Gangnam Severance Hospital and Peking Union Medical College Hospital, and DeepCatch X is supplied to seventeen branches of the Korea Association of Health Promotion.

That is standards support with named deployments behind it. No record system, PACS vendor or interface for returning results to the chart is named, and the integrations the company publicizes (NVIDIA Omniverse for 3D collaboration, AWS) are compute and rendering, not clinical workflow.

MEDIP PRO is stand alone software, according to its 510(k) summary, which places the studies it processes on the customer's own machines. For the other products no deployment option is stated, and the company publicizes an NVIDIA Omniverse connection and a relationship with AWS without saying which products use cloud processing or where. The company operates from Korea with offices in the United States and China and sells into all three, where rules on moving imaging data across borders differ sharply.

No price, tier or licensing model is published for any product, and inquiries go to sales contacts by region (Korea, the United States and China). The portfolio makes the gap wider: MEDIP PRO, the individual DeepCatch X modules, DeepFore, the ANATDEL 3D printing service and the anatomy education products are different propositions with different natural units, and nothing indicates whether they are licensed separately or bundled, or priced per study, per seat or per site. The 3D printing business is a per item service.

Coverage is broad and named, with validation behind part of it. Modalities span CT, MRI and chest radiography; clinical uses span cardiovascular, hepatic, pulmonary, infectious disease, neurosurgical and urological applications; buyers run from academic medical centers to medical and dental schools, device manufacturers and health screening operators, with DeepCatch X supplied to seventeen branches of the Korea Association of Health Promotion.

Validation stands behind parts of it: DeepFore in patients with chronic hepatitis B (Journal of Hepatology, 2025), MEDIP PRO's orbital segmentation in thyroid eye disease (PLOS ONE, 2026), and patient specific kidney models in robotic partial nephrectomy. MEDIP PRO's US clearance rested on bench testing with no clinical study.