Analytica chimica acta · 2026
Complementarity validation of strain identification based on LIBS spectra and gram-stained microscopic images.
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.
Yu Runheng, Liang Jiahui, Jiao Lusen, Wang Shuqing, Tian Zhihui, Chen Fei, Zhang Lei, Yin Wangbao, Xiao Liantuan, Jia Suotang
Аннотация
BACKGROUND: LIBS provides chemical fingerprints of strains, while Gram-stained microscopic images offer morphological traits including staining reaction, cell morphology, and spatial arrangement. These two modalities have distinct physical origins and potential complementarity, yet their combined value in complex multi-category strain identification remains to be verified.
RESULTS: Using 33 strains under a unified preparation workflow, we built single-modal classification models and three decision-level fusions (fixed-weighted probability, adaptive confidence, and Stacking meta-learning). LIBS single-modal (PCA-SVM on graphite) achieved Accuracy and Macro-F1 of 0.9694, with residual errors mainly on a few chemically similar strain pairs. Microscopic image single-modal (DN201) reached 0.9879, but still confused some morphologically similar pairs. Among fusions, adaptive confidence fusion (Fusion B) achieved the highest Accuracy and Macro-F1 values, both at 0.9991. The highest bidirectional mutual misclassification rate was 5.00% for LIBS, 6.00% for images, and dropped to 0.50% after Fusion B.
SIGNIFICANCE: These results indicate that, in the measurement-level evaluation under the current controlled cultivation and preparation conditions, LIBS chemical information and microscopic-image morphological information exhibit a certain degree of complementary discrimination. Their combination can improve classification performance on the current dataset and reduce some residual misclassifications of the single modalities, providing preliminary experimental evidence for subsequent research on multimodal microbial identification methods.
RESULTS: Using 33 strains under a unified preparation workflow, we built single-modal classification models and three decision-level fusions (fixed-weighted probability, adaptive confidence, and Stacking meta-learning). LIBS single-modal (PCA-SVM on graphite) achieved Accuracy and Macro-F1 of 0.9694, with residual errors mainly on a few chemically similar strain pairs. Microscopic image single-modal (DN201) reached 0.9879, but still confused some morphologically similar pairs. Among fusions, adaptive confidence fusion (Fusion B) achieved the highest Accuracy and Macro-F1 values, both at 0.9991. The highest bidirectional mutual misclassification rate was 5.00% for LIBS, 6.00% for images, and dropped to 0.50% after Fusion B.
SIGNIFICANCE: These results indicate that, in the measurement-level evaluation under the current controlled cultivation and preparation conditions, LIBS chemical information and microscopic-image morphological information exhibit a certain degree of complementary discrimination. Their combination can improve classification performance on the current dataset and reduce some residual misclassifications of the single modalities, providing preliminary experimental evidence for subsequent research on multimodal microbial identification methods.