Part 2: Seeing Muscle in Reat Time - CNNs and Ultrasound Body Compossition Asstssment

Ultrasound offers a portable and radiation‑free method for evaluating muscle and subcutaneous fat. Despite these advantages, ultrasound measurements have historically been operator dependent. Convolutional Neural Networks (CNNs) are a form of deep learning, and are now enabling automated interpretation of ultrasound images. This is making body composition analysis more accessible in clinical settings.

These CNN models analyze ultrasound textures and patterns to identify structures such as muscle fascia and fat layers. This allows automated measurement of muscle thickness, cross‑sectional area, and tissue quality.

Automated Muscle Thickness Measurement

CNN‑based ultrasound systems can automatically detect muscle boundaries and calculate muscle thickness without manual tracing. This capability is particularly valuable for monitoring sarcopenia, frailty, and rehabilitation progress.

Predicting Whole‑Body Composition

Researchers are exploring the use of ultrasound images from multiple anatomical sites combined with CNN models to estimate whole‑body fat mass and fat‑free mass. Early studies demonstrate promising agreement with laboratory methods such as air displacement plethysmography.

Clinical Example

ICU Muscle Monitoring Using CNN-Assisted Ultrasound

Loss of skeletal muscle mass is one of the most common and clinically significant complications in critically ill patients. Individuals admitted to an intensive care unit (ICU) often experience rapid muscle wasting, sometimes losing 10–20% of muscle mass within the first week of critical illness. Contributing factors include inflammation, immobilization, inadequate nutrition, and mechanical ventilation. Because muscle loss is strongly associated with prolonged recovery, ventilator dependence, and increased mortality, monitoring muscle status has become an important component of ICU care.

Ultrasound offers a practical method for assessing muscle mass at the bedside, but traditional ultrasound measurements require trained clinicians to manually identify muscle boundaries and measure thickness. This process can be time-consuming and subject to inter-observer variability. Convolutional neural networks (CNNs) are now being developed to automate this process, allowing clinicians to obtain rapid and reliable muscle measurements directly at the bedside.

Clinical Scenario

A 62-year-old patient is admitted to the ICU with severe pneumonia requiring mechanical ventilation. Because prolonged immobilization and systemic inflammation increase the risk of ICU-acquired weakness, the clinical team decides to monitor the patient’s muscle status throughout the hospital stay.

On the second day of admission, a clinician performs a bedside ultrasound scan of the quadriceps muscle, typically focusing on the rectus femoris and vastus intermedius muscles at the mid-thigh. The ultrasound probe is placed perpendicular to the skin surface, producing a cross-sectional image of the muscle.

Instead of manually tracing the muscle boundaries, the ultrasound image is immediately processed by a CNN-based analysis system integrated into the ultrasound device.

Step 1: Automatic Identification of Muscle Structures

The CNN analyzes the ultrasound image and identifies key anatomical structures within the scan. These include:

• the superficial fascia separating subcutaneous fat from muscle
• the muscle boundaries of the rectus femoris
• deeper muscle layers such as the vastus intermedius
• connective tissue structures

Because CNNs are trained on large datasets of labeled ultrasound images, they can reliably distinguish muscle tissue from surrounding structures despite the presence of ultrasound noise and speckle patterns.

Step 2: Automated Measurement of Muscle Thickness and Area

After identifying the relevant structures, the system automatically calculates several muscle metrics, including:

• rectus femoris muscle thickness
• combined quadriceps muscle thickness
• muscle cross-sectional area
• muscle echogenicity, which may indicate fat infiltration or muscle quality

These measurements are generated in real time, often within seconds, and displayed directly on the ultrasound screen or transmitted to the electronic health record.

Step 3: Longitudinal Monitoring of Muscle Loss

The patient undergoes repeat ultrasound scans every 2–3 days during the ICU stay. Because the CNN performs the measurements automatically, the process requires minimal additional time for the clinical staff.

Over the course of one week, the system detects a progressive decline in quadriceps muscle thickness, suggesting rapid muscle loss associated with immobilization and critical illness. The automated system generates a trend graph showing changes in muscle thickness over time. This visualization allows the clinical team to quickly recognize the patient’s worsening muscle status.

Step 4: Clinical Intervention

The detection of rapid muscle loss prompts a multidisciplinary response. The ICU team may implement several interventions, including:

• nutritional optimization, such as increasing protein intake or adjusting enteral feeding strategies
• early mobilization and physiotherapy, including passive or active limb movement
• evaluation of sedation strategies to allow more patient movement
• monitoring for ICU-acquired weakness

By identifying muscle loss early, clinicians can intervene before severe muscle wasting occurs, potentially improving recovery and functional outcomes.

Why CNN-Assisted Ultrasound Is Valuable in the ICU

Several features make this technology particularly useful in critical care settings:

Rapid bedside assessment- Muscle measurements can be obtained in seconds without transporting the patient.

Reduced operator dependence- Automated segmentation minimizes variability between clinicians.

Frequent monitoring- Muscle status can be tracked repeatedly without radiation exposure.

Integration with patient monitoring systems- Automated measurements can be stored in the electronic health record and analyzed alongside other clinical data.

Broader Implications

CNN-assisted ultrasound represents a major step toward real-time monitoring of body composition in critically ill patients. Because ultrasound devices are portable and widely available, AI-based analysis could allow routine muscle monitoring in ICUs worldwide.

As these systems continue to evolve, they may also support:

• automated detection of sarcopenia or frailty
• prediction of ventilator weaning success
• monitoring of rehabilitation progress after ICU discharge

Ultimately, integrating AI-driven ultrasound analysis into critical care workflows could help clinicians better understand and manage the profound metabolic changes that occur during severe illness.

More Reading -- Key References

He, K., Li, Y., Zhang, X., Chen, Y., & colleagues. (2024). Deep learning prediction of body composition from ultrasound images. Sensors, 24(3), 987.

Behboodi B.  Deep learning frameworks for automated muscle thickness detection using ultrasound imaging. IEEE Access, 13, 11845–11857.

Ardhianto, P., Lee, J., & Kim, H. (2021). Deep learning applications in muscle ultrasound imaging: Challenges and opportunities. Applied Sciences, 11(9), 4021.

Mustapoevich, D., Shpanskaya, K., Sadykova, G., & colleagues. (2023).
Artificial intelligence applications in sarcopenia detection and body composition analysis: Current evidence and future directions. Healthcare, 11(18), 2483.

Kolarik, M., Burget, R., Uher, V., Dutta, M. K., & others. (2023).  Explainability of deep learning models in medical imaging: A survey. PeerJ Computer Science, 9, e1253.