Hirofumi Kobayashi

Title: Hirofumi Kobayashi and Machine-Learning Methods for Immunostain Analysis.

Introduction

Hirofumi Kobayashi is an inventor associated with Pictor Labs, Inc., in Los Angeles, California. His patent work addresses the refinement of image annotations used to train machine-learning systems on noisy or inaccurate immunostaining data.

Latest Patents

Kobayashi is associated with 1 patent, titled “Systems and Methods for Training Machine Learning on Noisy and Inaccurate Immunostains.” The invention describes a method for refining virtually stained image annotations. It separates an image of stained tissue into counterstain and immunostaining visual-signal images using a linear transformation. It then identifies and segments nuclei in the counterstain image, creating single-cell segments in a refined counterstain image.

The method compares immunostaining signals within those segments against a user-defined threshold to identify true-positive cells. It refines the immunostaining signals based on protein localization or tissue morphology and the identified true-positive cells. Finally, it overlays the counterstain image with the refined immunostaining image to produce an improved virtually stained image annotation.

Career Highlights

Kobayashi’s listed patent connects his work with computational pathology and machine-learning training methods. By refining annotations derived from tissue images, the approach is designed to address inaccuracies in immunostaining data used in image analysis.

Collaborations

Kobayashi is associated with Pictor Labs, Inc. His listed coworkers include Yair Rivenson and Luis Carvalho.

Conclusion

Hirofumi Kobayashi’s patent focuses on improving virtually stained tissue-image annotations, supporting more reliable machine-learning analysis of immunostaining data.