Classification of changes occurring in lung patient during radiotherapy using relative γ analysis and hidden Markov models.

PURPOSE: To present a new automated patient classification method based on relative gamma analysis and hidden Markov models (HMM) to identify patients undergoing important anatomical changes during radiation therapy. METHODS: Daily EPID images of every treatment field were acquired for 52 patients treated for lung cancer. In addition, CBCT were acquired on a regular basis. Gamma analysis was performed relative to the first fraction given that no significant anatomical change was observed on the CBCT of the first fraction compared to the planning CT. Several parameters were extracted from the gamma analysis (e.g., average gamma value, standard deviation, percent above 1). These parameters formed patient-specific time series. Data from the first 24 patients were used as a training set for the HMM. The trained HMM was then applied to the remaining 28 patients and compared to manual clinical evaluation and fixed thresholds. RESULTS: A three-category system was used for patient classification ranging from minor deviations (category 1) to severe deviations (category 3) from the treatment plan. Patient classified using the HMM lead to the same result as the classification made by a human expert 83% of the time. The HMM overestimate the category 10% of the time and underestimate 7% of the time. Both methods never disagree by more than one category. In addition, the information provided by the HMM is richer than the simple threshold-based approach. HMM provides information on the likelihood that a patient will improve or deteriorate as well as the expected time the patient will remain in that state. CONCLUSION: We showed a method to classify patients during the course of radiotherapy based on relative changes in EPID images and a hidden Markov model. Information obtained through this automated classification can complement the clinical information collected during treatment and help identify patients in need of a plan adaptation.

Characterization of lung tumors motion baseline using cone-beam computed tomography.

PURPOSE: To characterize the interfractional variability in lung tumor volume, position, and tumor boundaries. METHODS: Cone-beam computed tomography (CBCT) scans were acquired weekly during the course of treatment for 34 lung cancer patients (1-20 scans) with large tumors. Spatial registration based on bones was performed between contoured planning CT and CBCT. Gross tumor volume (GTV) on each CBCT was then contoured. Tumor volume, centroid, and boundaries variability were quantified. A commercial deformable registration software was tested and results were compared to manual contours. RESULTS: Mean volume reduction was 41 ± 32% (p < 0.001) after an average time of 51 days. Tumor centroid drifts were 0.03, 0.14, and -0.13 cm in right-left (RL), anterior-posterior (AP), and superior-inferior (SI) directions with standard deviations of 0.55, 0.50, and 0.51 cm. GTV boundaries displacements were -0.27, -0.14, and -0.16 cm with standard deviations of 0.64, 0.57, and 0.59 cm in RL, AP, and SI directions. Relative error between deformed and manual contours with the commercial deformable registration software rose up exponentially with the GTV decrease. CONCLUSIONS: GTV size changes for large lung tumors are similar to those for standard tumors. Magnitude absolute values of displacement vector for centroid and boundaries shifts show that there is not a preferred direction for the drifts but shrinkage.