X-ray CT AI apps everywhere

CT Image Generation Platform for AI/ML Applications

To address the trade-off between image quality enhancement and radiation dose reduction in X-ray CT systems, I contributed to the research and development of next-generation AI image generation technology which realized the simultaneous achievement of improved diagnostic accuracy and reduced patient burden in clinical settings. Furthermore, while adding new functionality to medical devices already released in the market is considered challenging due to regulatory requirements and system resource limitations, I successfully accomplished an integrated platform combining scalable cloud-edge computing with medical diagnostic equipment, obtained approval from regulatory authorities including the FDA, and delivered comprehensive technology solutions to healthcare facilities worldwide.

Cloud-Edge Computing Architecture

To enable the use of new functionality across various X-ray CT systems, I contributed to build a microservices architecture utilizing Docker and Kubernetes. By combining this with a distributed data processing and monitoring system, I achieved real-time AI inference. In CT Smart Subscription, I led to implement a hybrid architecture that integrates edge computing with cloud orchestration, enabling customers to benefit from AI technology regardless of hospital size or equipment configuration (Smart Subscription for CT, 2025). See the below videos as reference.

AI-Enhanced Image Generation

In True Enhance DL, a virtual monochromatic X-ray imaging technology, I contributed to adopt a Residual Dense Block architecture and developed a method to generate high-contrast images from monochromatic X-ray CT images. This technology made it possible to reduce contrast agent usage while maintaining the image quality necessary for diagnosis, without requiring special equipment such as Dual Energy CT (Effortless Recon DL Portfolio, 2024; Revolution Ascend Platform, 2024).

Additionally, in True Fidelity DL, a noise reduction technology, I contributed to adopt a Fully Convolutional Networks (FCN) architecture and achieved a method that selectively removes noise generated during low-dose imaging while preserving image sharpness and texture. This technology enables significant reduction in patient radiation exposure while improving diagnostic quality (Effortless Recon DL Portfolio, 2024; Revolution Ascend Platform, 2024).

Both technologies obtained FDA approval and achieved commercial deployment in global markets. These technologies have been proven to reduce radiation exposure by up to 30% while maintaining the image quality necessary for diagnosis (Deep Learning Image Reconstruction (K230807), 2023; True Enhance DL (K233698), 2024).

Runtime-Agnostic AI/ML Framework

To achieve efficient AI inference in heterogeneous computing environments, I led to develop a runtime-independent solution integrating CUDA/CuDNN, OneAPI/DNN, ONNX, and PyTorch/TensorFlow. This design made it possible to execute AI applications with optimal performance regardless of hardware configuration.

I also led to achieve more than 2x acceleration in AI inference processing through quantization techniques and multi-GPU utilization, and by combining this with low-level optimization using CUDA and C++, we’ve met the real-time processing requirements demanded in clinical settings.