Analytica chimica acta · 2026
Dual-channel CNN fusion of raw and preprocessed serum Raman spectra for robust echinococcosis screening.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Chen Haoran, Zheng Xiangxiang, Huang Jiahui, Huang Yukang, Xu Liang, Wu Guohua, Lü Guodong
Аннотация
BACKGROUND: Serum Raman spectroscopy provides label-free molecular fingerprints for bioanalytical screening, but conventional models often rely on a single preprocessing pipeline. Because preprocessing can alter spectral distributions and attenuate information retained in raw spectra, this study developed a dual-channel fusion strategy for echinococcosis screening.
RESULTS: We analyzed 573 serum samples from hepatic echinococcosis (HE), liver cirrhosis (LC), hepatocellular carcinoma (HCC), and normal controls (NC); 458 formed the development set and 115 were reserved for held-out testing. We developed a dual-channel convolutional neural network (DCNN) that fused complementary features from raw and preprocessed Raman spectra. Eight preprocessing pipelines and three fusion strategies were evaluated using identical stratified five-fold splits within the development set, and the configuration with the highest mean five-fold accuracy was fixed before held-out testing. The selected fusion strategy was then assessed across all eight pipelines against single-channel models. Under the locked configuration, the DCNN achieved 94.98% ± 1.62% mean accuracy, compared with 89.95% ± 4.36% for the preprocessed-spectrum CNN (P-CNN) and 88.42% ± 2.66% for the raw-spectrum CNN (R-CNN). Held-out accuracies were 95.65% for DCNN, 91.30% for P-CNN, and 89.57% for R-CNN. For binary screening, a DCNN trained on the development set achieved 96.52% accuracy on the held-out set, whereas the dot immunogold filtration assay (DIGFA) yielded 88.70% on the same samples.
SIGNIFICANCE: Feature-level fusion improved the use of complementary spectral information and reduced dependence on a single preprocessing pipeline. This strategy provides a robust framework for rapid, label-free serum Raman screening of echinococcosis.
RESULTS: We analyzed 573 serum samples from hepatic echinococcosis (HE), liver cirrhosis (LC), hepatocellular carcinoma (HCC), and normal controls (NC); 458 formed the development set and 115 were reserved for held-out testing. We developed a dual-channel convolutional neural network (DCNN) that fused complementary features from raw and preprocessed Raman spectra. Eight preprocessing pipelines and three fusion strategies were evaluated using identical stratified five-fold splits within the development set, and the configuration with the highest mean five-fold accuracy was fixed before held-out testing. The selected fusion strategy was then assessed across all eight pipelines against single-channel models. Under the locked configuration, the DCNN achieved 94.98% ± 1.62% mean accuracy, compared with 89.95% ± 4.36% for the preprocessed-spectrum CNN (P-CNN) and 88.42% ± 2.66% for the raw-spectrum CNN (R-CNN). Held-out accuracies were 95.65% for DCNN, 91.30% for P-CNN, and 89.57% for R-CNN. For binary screening, a DCNN trained on the development set achieved 96.52% accuracy on the held-out set, whereas the dot immunogold filtration assay (DIGFA) yielded 88.70% on the same samples.
SIGNIFICANCE: Feature-level fusion improved the use of complementary spectral information and reduced dependence on a single preprocessing pipeline. This strategy provides a robust framework for rapid, label-free serum Raman screening of echinococcosis.