AI Tracks Human Diaphragm Movement with 88% Accuracy

Introduction

A research team from Taipei Medical University and the National Taipei University of Technology has successfully tracked the human diaphragm’s movement with an accuracy of 88% using a new artificial intelligence system. Their work, published in Computers in Biology and Medicine, demonstrates how controlled breathing patterns can be measured with precision, offering a new window into the physiology of breath control and its applications in modern medicine.

AI Measures Diaphragm Motion with High Precision

Led by corresponding author H.C. Chuang, the researchers developed a non-invasive system to track respiration in real time. They used an ultrasound probe to image the diaphragm—the primary muscle driving breath—and fed this data into a deep learning algorithm called SE-ATT-YOLO. This AI was trained to identify the diaphragm and create a digital “bounding box” around it, tracking its coordinates frame-by-frame as it moved with each breath.

The system’s performance was quantified with a metric called mean average precision (mAP), where 1.0 represents perfect detection. Their model achieved an mAP of 0.88, outperforming the previous standard model’s score of 0.85. It processed images at about 50 frames per second, confirming its ability to operate in real time. This level of precision allows for a detailed, objective analysis of how the diaphragm behaves during different breathing maneuvers, moving beyond subjective sensation to measurable biomechanics.

Slow, Controlled Breathing Yields the Most Stable Signal

To test the system’s accuracy in tracking different breathing patterns, the team compared its real-time readings against pre-recorded respiratory signals. They calculated the Root Mean Square Error (RMSE)—a measure of tracking error where a lower number means greater accuracy—for four distinct patterns.

The results were telling. A simulated “baseline shift” in breathing had an RMSE of 4.342. A regular “sinusoidal signal,” mimicking normal quiet breathing, scored 3.105. A single “deep breath” produced an error of 1.778. Most significantly, a “slow signal” designed to imitate paced, controlled breathing achieved the lowest error of all: 1.667. This finding indicates that slow, deliberate breathing creates the most predictable and easily tracked movement of the diaphragm. The steadier the breath, the more consistently the AI—and by extension, a medical device it controls—can predict and follow its motion.

From Radiation Oncology to Breathwork Science

The study’s primary aim was advancing cancer treatment. In radiotherapy for lung, liver, or pancreatic tumors, even a few millimeters of movement caused by normal breathing can cause radiation to miss the tumor and damage healthy tissue. This new Respiratory Motion Compensation System (RMCS), driven by the AI’s tracking, can adjust the radiation beam in sync with the patient’s breath, targeting the tumor more precisely. This application is explored in our related article on AI enhancing controlled breathing for better radiotherapy.

Beyond oncology, the research provides a powerful tool for respiratory science. By offering a validated method to quantify diaphragm movement, it creates a bridge between ancient breathwork practices and modern physiology. Techniques like box breathing, a cornerstone of pranayama and modern tactical breathing, involve consciously slowing and regulating the breath. This study’s data suggests such practices produce a uniquely stable diaphragmatic signal. This measurable stability correlates with the physiological calm—reduced heart rate and blood pressure—associated with these practices, which are detailed in our analysis of pranayama benefits for heart health.

Practical Implications for Health and Performance

For patients undergoing thoracic or abdominal radiotherapy, this technology means treatments can be both more effective and safer. Clinics may increasingly use guided breathing protocols to help patients achieve the slow, consistent pattern that the system tracks best, directly improving outcomes.

For individuals interested in breathwork, the study reinforces a key principle: slow control matters. The low error rate for the “slow signal” pattern provides a mechanistic explanation for why disciplined practices like box breathing are effective. They train the respiratory system to produce a stable, repeatable output, which the nervous system interprets as a signal of safety, thereby downregulating the anxiety cycle. While the study did not measure heart rate variability or stress hormones, the biomechanical stability it captured is a likely precursor to those well-documented autonomic benefits.

A limitation of the work is that it was a technical validation using recorded signals. Further research is needed to confirm its performance across a wide population during live, patient-guided breathwork sessions.

Conclusion

The Taipei-based research demonstrates that advanced AI can map human breathing with high accuracy. Its immediate use in radiotherapy highlights a life-saving application of breath control. More broadly, it provides a scientific lens through which to view the deliberate, steady patterns of therapeutic breathwork, linking their physical mechanics to their profound effects on health and resilience.

Sources:
https://pubmed.ncbi.nlm.nih.gov/40544802/
https://pubmed.ncbi.nlm.nih.gov/40269708/
https://pubmed.ncbi.nlm.nih.gov/40253678/