Ikematsu Hiroaki, Takara Yohei, Nishihara Keiichiro, Kano Yuki, Owaki Yuji, Okamoto Ryuji, Fujiwara Takahisa, Takamatsu Toshihiro, Yamada Masayuki, Tomioka Yutaka, Takeshita Nobuyoshi, Inaba Atsushi, Sunakawa Hironori, Nakajo Keiichiro, Murano Tatsuro, Kadota Tomohiro, Shinmura Kensuke, Koga Yoshikatsu, Yano Tomonori
Division of Science and Technology for Endoscopy, Exploratory Oncology Research and Clinical Trial Center, National Cancer Center, Kashiwa, Japan.
Department of Gastroenterology, IMSUT Hospital, The Institute of Medical Science, The University of Tokyo, 4-6-1 Shirokanedai, Minato-ku, Tokyo, 108-8639, Japan.
J Gastroenterol. 2025 Jan;60(1):77-85. doi: 10.1007/s00535-024-02163-2. Epub 2024 Oct 23.
Fecal immunochemical tests are commonly performed for colorectal cancer screening. Instant fecal occult blood measurement in toilet bowel movements would improve convenience. Hyperspectral imaging (HSI) enables the nondestructive evaluation of materials that are difficult to assess visually. This study aimed to determine whether HSI could be used to identify fecal occult blood on stool surfaces.
The study included 100 patients who underwent colonoscopy, divided into groups A and B (50 patients, each) for creating a discriminant algorithm and validating the accuracy of the algorithm, respectively. In group A, 100 areas were randomly selected from the stool surface, and the fecal occult blood quantitative values were measured and photographed using a hyperspectral camera (cutoff: > 400 ng/mL). A discriminant algorithm image was created to extract spectral feature differences obtained from HSI via machine learning. In group B, 250 random areas were evaluated and compared to fecal occult blood quantitative values, measuring sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV).
Groups A and B comprised 28 and 26 patients with cancer, respectively. Cancer detection sensitivity at the 400 ng/mL cutoff was 67.9% and 42.3% in groups A and B, respectively. The discriminant algorithm image exhibited high accuracy in group A (sensitivity; 77.1%, specificity; 96.9%, accuracy; 90.0%, PPV; 93.1%, NPV; 88.7%). In group B, the sensitivity, specificity, accuracy, PPV, and NPV were 83.3, 92.9, 90.8, 76.3, and 95.3%, respectively.
HSI can effectively discriminate high quantitative fecal occult blood, highlighting its potential for improved colorectal cancer screening.
粪便免疫化学检测常用于结直肠癌筛查。在马桶排便时即时检测粪便潜血将提高便利性。高光谱成像（HSI）能够对难以通过视觉评估的材料进行无损评估。本研究旨在确定HSI是否可用于识别粪便表面的潜血。
该研究纳入了100例行结肠镜检查的患者，分为A组和B组（每组50例），分别用于创建判别算法和验证算法的准确性。在A组中，从粪便表面随机选择100个区域，使用高光谱相机测量并拍摄粪便潜血定量值（临界值：>400 ng/mL）。创建判别算法图像，通过机器学习提取从HSI获得的光谱特征差异。在B组中，对250个随机区域进行评估，并与粪便潜血定量值进行比较，测量灵敏度、特异性、准确性、阳性预测值（PPV）和阴性预测值（NPV）。
A组和B组分别有28例和26例癌症患者。在400 ng/mL临界值时，A组和B组的癌症检测灵敏度分别为67.9%和42.3%。判别算法图像在A组中表现出较高的准确性（灵敏度；77.1%，特异性；96.9%，准确性；90.0%，PPV；93.1%，NPV；88.7%）。在B组中，灵敏度、特异性、准确性、PPV和NPV分别为83.3%、92.9%、90.8%、76.3%和95.3%。
HSI能够有效鉴别高定量的粪便潜血，凸显了其在改善结直肠癌筛查方面的潜力。
