Multi-component exercise intervention methods for intelligent assisted diagnosis of sarcopenia in the elderly based on deep learning

Population aging has made sarcopenia a major public health challenge in geriatric medicine. Current clinical diagnosis relies on large specialized equipment, resulting in low screening efficiency and limited capacity for early risk prediction. In addition, conventional intervention strategies follow standardized protocols and cannot accommodate individual differences in muscle function among older adults. Consequently, diagnosis and intervention have remained disconnected, creating a persistent research challenge. To address these limitations, this study integrated multimodal data, including muscle ultrasound images, body composition indices, muscle strength measurements, and longitudinal clinical follow-up records. An integrated framework for intelligent sarcopenia diagnosis and personalized multicomponent exercise intervention was developed based on an improved Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model. Z-score normalization was applied to eliminate scale differences among heterogeneous data sources. The CNN extracted spatial features from muscle ultrasound images, whereas the LSTM captured temporal patterns from longitudinal follow-up data. Together, these components enabled early identification of sarcopenia and accurate risk stratification. Experimental results demonstrated that the proposed CNN-LSTM hybrid model achieved a diagnostic accuracy of 92.3%, with an area under the receiver operating characteristic curve of 0.95. The recall rate for mild sarcopenia reached 89.5%, and the overall performance significantly exceeded that of single-modality models and conventional machine learning algorithms. After a 24-week personalized exercise intervention, participants exhibited a 12.7% increase in total muscle mass. Grip strength and knee extensor strength increased by 18.2% and 15.5%, respectively, while walking speed improved by 0.22 m/s. The incidence of intervention-related adverse events was only 3.3%, and all outcome measures outperformed those achieved with the standardized intervention program. The proposed closed-loop diagnosis-intervention framework extended the application of deep learning to age-related degenerative diseases. It also provided a low-cost and practical solution for large-scale sarcopenia screening and precision health management in primary healthcare settings, offering substantial practical value for advancing precision-oriented geriatric healthcare.