An architecture for Deep Learning based automatic bioimages segmentation for sarcopenia evaluation

Sarcopenia is a clinical condition marked by loss of muscle mass and strength, leading to reduced mobility and quality of life. Accurate identification and quantification of muscle mass are essential for the timely diagnosis and treatment of sarcopenia. To calculate muscle volumes and sarcopenia indexes, segmentation techniques are required on CT images, helping to adjust treatments for chronic diseases. However, evaluating muscle volumes and thus determining sarcopenia indexes currently is highly dependent on manual image segmentation by human operators. We propose a deep learning architecture for automatic muscle mass segmentation, integrating DeepLabv3+ and U-Net3+ models for 2D and 3D segmentation, respectively. These models have been tested on available datasets and combined using an ensemble learning approach to enhance predictive accuracy. This proposed architecture can be integrated into clinical workflows for the assessment of sarcopenia, increasing the reliability and efficiency of image-based diagnoses.