Quantitative early decision making metric for identifying irregular breathing in 4DCT




To develop a quantitative early decision making metric for prediction of breathing pattern and irregular breathing and validate the metric in a large patient population receiving clinical phase-sorted four-dimensional computed tomography (4DCT).


This study employed three patient cohorts. The first cohort contained 47 patients, imaged with a nonclinical tidal volume metric. The second cohort contained a sample of 256 patients who received a clinical 4DCT. The third cohort contained 86 patients who received three 4DCT scans at 1-week increment during the course of radiotherapy. The second and third cohorts did not have tidal volume measurements, as per standard radiation oncology clinical practice. Based on a previously published technique that used a single abdominal surrogate, the ratio of extreme inhalation tidal volume to normal inhalation tidal volume (κ) metric was calculated and the patient breathing pattern was characterized. The use of a single surrogate precluded the use of a κ determined by tidal volume, so a κrel was defined based on the amplitude of the surrogate. Patients were classified as either Type 1 or Type 2, based on a previously published technique, where Type 1 patients were apneic at end of exhalation and Type 2 patients exhibited forced respiration. The Ansari–Bradley test was used to determine the statistical similarity between the Type 1 and Type 2 distributions. A Kruskal–Wallis one way analysis of variance was used to determine the statistical similarities among the classified breathing types, κrel, and the qualified medical physicist denoted breathing classification (regular or irregular). Receiver operator characteristic curves were used to quantitatively determine optimal cutoff value jκ and efficiency cutoff value τκκrel to provide a quantitative early warning of irregular breathing during 4DCT procedures.


The statistical tests show a significant consistency for the breathing pattern classifications between the physiologically measured cohort #1 and the remaining cohorts. The classification types were statistically different between Type 1 and Type 2 patients over all cohorts. Values of κrel in excess of 1.72 indicated a substantial presence of irregular breathing that could negatively affect the quality of a 4DCT image dataset. Values of κrel in lower than 1.45 indicated minimal presence of irregular breathing. For values of κrel such that jκκrelτκ, the decision to reacquire the 4DCT would be at the discretion of the physician. This accounted for only 11.9% of the patients in this study. The magnitude of κrel held consistent over three weeks of treatment for 73% of the patients in cohort #3.


The decision making metric based on κ was shown to be an accurate classifier of regular and irregular breathing patterns in a large patient population. Breathing type, as defined in a previous published work, was accurately classified by κrel with the use of a single respiratory surrogate compared to the physiological use of multiple respiratory surrogates. This work provided a quantitative early decision making metric to quickly and accurately assess breathing patterns as well as the presence and magnitude of irregular breathing during 4DCT.