Park Sang Mok, Kwon Semin, Ji Yuhyun, Sakthivel Haripriya, Leem Jung Woo, Kwak Yunsang, Huang Jonathan, Chiu George T-C, O'Brien Andrew R, Konger Raymond L, Wang Ying, Kim Young L
Weldon School of Biomedical Engineering, Purdue University, West Lafayette, IN 47907, USA.
Department of Mechanical System Engineering, Kumoh National Institute of Technology, 61 Daehak-ro, Gumi-si, Gyeongsangbuk-do 39177, Republic of Korea.
Sci Adv. 2025 Jun 6;11(23):eadt4831. doi: 10.1126/sciadv.adt4831. Epub 2025 Jun 4.
Despite advances in machine learning and computer vision for biomedical imaging, machine reading and learning of colors remain underexplored. Color consistency in computer vision, color constancy in human perception, and color accuracy in biomedical imaging are intertwined, complicating digital color-based diagnostics. Existing color reference charts and correction algorithms are inadequate for mobile health (mHealth) and telemedicine in digital health applications where detecting subtle color changes is critical. We present a machine reading and learning platform for color recognition and quantification to extract diagnostic information from colors. A unique combination of spectroscopic gamut determination, reference color optimization, nonsubjective quantification metrics, and neural network-based color recovery retrieves absolute colors of biological tissue. Studies on inflammation bioimaging of photocarcinogenesis and mHealth blood hemoglobin assessment demonstrate accuracy and precision in color recovery across diverse acquisition scenarios. The reported framework overcomes limitations of conventional colorimetric detection, enabling machine-compatible color-based bioassays and bioimaging, advancing digital diagnostics.
尽管机器学习和计算机视觉在生物医学成像方面取得了进展，但机器对颜色的读取和学习仍未得到充分探索。计算机视觉中的颜色一致性、人类感知中的颜色恒常性以及生物医学成像中的颜色准确性相互交织，使得基于数字颜色的诊断变得复杂。现有的颜色参考图表和校正算法在移动健康（mHealth）和远程医疗等数字健康应用中并不适用，因为在这些应用中检测细微的颜色变化至关重要。我们提出了一个用于颜色识别和量化的机器读取和学习平台，以从颜色中提取诊断信息。光谱色域确定、参考颜色优化、非主观量化指标和基于神经网络的颜色恢复的独特组合可检索生物组织的绝对颜色。对光致癌炎症生物成像和移动健康血液血红蛋白评估的研究证明了在不同采集场景下颜色恢复的准确性和精确性。所报道的框架克服了传统比色检测的局限性，实现了与机器兼容的基于颜色的生物测定和生物成像，推动了数字诊断的发展。
