Summary

  • SK Telecom launched X Caliber in September 2022, an AI-based diagnostic support service developed using medical imaging data from national veterinary schools to address growing demand for pet X-ray diagnostics.
  • In 2025, the inference infrastructure for X Caliber will be fully migrated from GPUs to Rebellions’ ATOM™ NPU.
  • This is the first large-scale subscription-based AI service used by approximately one thousand veterinary clinics nationwide that operates entirely on an NPU-only infrastructure.

Challenge

As the population of companion animals grows, the demand for X-ray image interpretation in veterinary clinics is rapidly increasing. However, Korea has only a few hundred board-certified veterinary radiologists. Most clinics must rely on general veterinarians to interpret images themselves, which creates a significant workload and potential for diagnostic variability. To address this, SK Telecom partnered with major national veterinary schools to develop X Caliber, an AI-assisted X-ray diagnostic service built on high-quality medical datasets.

X Caliber is the first service in Korea’s veterinary imaging domain to receive approval from the Animal and Plant Quarantine Agency as an animal medical device. It identifies abnormalities in musculoskeletal, thoracic, and abdominal X-ray images for dogs and cats. It also provides automatic heart size measurements from thoracic X-rays, helping veterinarians reduce interpretation time and improve workflow efficiency.

Clinical trials showed up to 97 percent agreement with specialist interpretation, demonstrating high accuracy. However, providing high-quality AI diagnostics to clinics nationwide around the clock requires a highly efficient inference infrastructure.

Solution

SK Telecom re-designed the inference architecture for X Caliber with power efficiency as the priority. To resolve the high power consumption and cost of GPU-based servers, the company adopted Rebellions’ ATOM™ NPU as the core of the new inference environment. ATOM™ NPU delivers image-inference performance comparable to GPU servers while significantly reducing power consumption, making it more cost-effective for long-term operation. By transitioning to NPU-based inference, SK Telecom improved operational efficiency and established a stable infrastructure capable of handling real-time requests from veterinary clinics across the country without GPU resources.

Result

After validating the stability and performance of the ATOM™ NPU through a proof-of-concept phase, SK Telecom fully replaced the GPU-based inference infrastructure of X Caliber with an NPU-only environment. X Caliber is delivered as a cloud-based web service and is currently used in approximately one thousand veterinary clinics nationwide. This deployment demonstrates the practical viability of Korean AI processors for medical imaging analysis in real-world clinical environments. Building on its demonstrated success in the domestic market, X Caliber is now expanding into overseas markets.

Appendix: NPU Usage Guide

Get started now: Prepare a pretrained model → Compile with the RBLN Compiler → Load the compiled model and run inference