---------------------------------------------------------------------------------------------------
Comming Soon
---------------------------------------------------------------------------------------------------
Comming Soon
---------------------------------------------------------------------------------------------------
Company project in South Korea | Computer vision Based Real Time Sasang Constitution Types Classification and Deploy on webserver using Flask.
Key Achievements:
Developed a real-time computer vision model for Sasang constitution type classification.
Successfully deployed the AI model on a web server using Flask for interactive access.
Optimized the system for efficient inference and user-friendly performance.
Demonstrated practical application of AI in traditional medicine analysis.
Korean Traditional Medication Real time.
---------------------------------------------------------------------------------------------------
We developed an efficient AI framework (WSI-WDM) for synthesizing histopathology whole slide images (WSIs) to support automated cancer detection. Gigapixel WSIs are divided into smaller patches, and wavelet transforms are applied at both the image and feature levels, reducing resolution and significantly speeding up training and inference.
Key achievements include:
Creating a scalable WSI synthesis framework for high-dimensional medical images;
Ensuring high-quality generated data using metrics such as FID, Recall, and inference speed;
Demonstrating faster and more accurate performance compared to existing diffusion-based methods.
This approach enables the development of robust AI models that can assist pathologists in diagnosing cancer faster and more accurately, paving the way for real-world clinical AI applications.
---------------------------------------------------------------------------------------------------
In this study, we propose PathoLDM, an innovative two-stage framework for generating non-melanoma skin cancer histopathology images. In the first stage, input images are encoded into a compact latent representation using a trained autoencoder. These latent features are then used in a diffusion-based process, preserving semantic and structural details while improving computational efficiency by avoiding direct manipulation of high-dimensional pixel data.
Key contributions include:
Development of the PathoLDM two-stage model combining an autoencoder and latent-space diffusion for histopathology image generation.
Preservation of fine-grained textures and anatomical structures, critical for accurate cancer characterization and diagnosis.
Comprehensive evaluation using FID, Inception Score (IS), and inference time to assess image fidelity, diversity, and efficiency.
Comparative analysis demonstrating superior performance over DD-GAN and Wavelet Diffusion Models (WDM) in both image quality and computational efficiency
---------------------------------------------------------------------------------------------------
Comming on the way
---------------------------------------------------------------------------------------------------
Interview1.pptx