A reliable contour detection method for volatomics analysis with comprehensive two-dimensional gas chromatography...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Data Science

Background:

• Comprehensive Two-Dimensional Gas Chromatography (GC×GC) offers high resolution for complex volatile analyses.
• Existing contour detection methods struggle with overlapped contours, leading to false negatives and inaccurate quantification in GC×GC data.
• Manual curation of GC×GC data is time-consuming and prone to errors.

Purpose of the Study:

• To develop a hybrid deep learning framework to overcome false negatives caused by overlapped contours in GC×GC analysis.
• To improve the accuracy and efficiency of automated contour detection and quantification in complex GC×GC datasets.
• To reduce the need for manual intervention in processing GC×GC data.

Main Methods:

• A hybrid deep learning framework integrating image classification (ResNet18) and instance segmentation (YOLO 11l) was developed.
• Initial contour maps were generated using an improved PeakCET v2 with a Laplacian operator.
• ResNet18 classified single and multi-peak contours, while YOLO 11l segmented overlapping contours.

Main Results:

• ResNet18 achieved 98.59% classification accuracy for contours.
• YOLO 11l demonstrated strong segmentation performance with mAP50 > 87% and the highest mAP50-95.
• The hybrid model showed generalizability and robustness on diverse rose oil datasets.

Conclusions:

• The proposed ResNet18-YOLO 11l pipeline effectively reduces false negatives in GC×GC analysis caused by overlapped contours.
• This automated approach offers a time-efficient solution for processing complex GC×GC data, minimizing manual curation.
• The framework enhances the reliability and throughput of quantitative analyses in metabolomics and other volatile profiling applications.