Assessing individual aging has always been an important topic in aging research. Caenorhabditis elegans (C. elegans) has a short lifespan and is a popular model organism widely utilized in aging research. Studying the differences in C. elegans life stages is of great significance for human health and aging. In order to study the differences in C. elegans lifespan stages, the classification of lifespan stages is the first task to be performed. In the past, biomarkers and physiological changes captured with imaging were commonly used to assess aging in isogenic C. elegans individuals. However, all of the current research has focused only on physiological changes or biomarkers for the assessment of aging, which affects the accuracy of assessment. In this paper, we combine two types of features for the assessment of lifespan stages to improve assessment accuracy. To fuse the two types of features, an improved high-efficiency network (Att-EfficientNet) is proposed. In the new EfficientNet, attention mechanisms are introduced so that accuracy can be further improved. In addition, in contrast to previous research, which divided the lifespan into three stages, we divide the lifespan into six stages. We compared the classification method with other CNN-based methods as well as other classic machine learning methods. The results indicate that the classification method has a higher accuracy rate (72%) than other CNN-based methods and some machine learning methods.

Caenorhabditis elegans (C. elegans) has a short life span and is a popular and excellent model for aging studies. The life stage prediction of C. elegans is of great significance to the studies of antiaging drug screening and genetic screening. Currently, most of the existing researches based on life prediction of C. elegans only focus on extracting features from physiological changes, lacking the possibility of technological migration, which has limited significance for human research. We combined the fusion protein aggregation in C.elegans with deep learning, and proposed a classification method of C. elegans life stage that combines the target detection method YOLOv3 with a bilinear fusion network. As the experimental results indicate, the proposed framework can achieve the 75.33% prediction accuracy, illustrating that it is a promising tool, has a higher possibility of technology migration, and has potential value for future research on human lifespan.