Automatic selection of lung cancer patients for adaptive radiotherapy using cone-beam CT imaging

Abstract

Background and purpose: Anatomical changes in non-small cell lung cancer (NSCLC) patients may lead to unacceptable treatment results, requiring adaptive radiotherapy (ART). The method proposed in this study describes the proof-of-principle to automatically select patients eligible for ART. Materials and methods: A method was developed flagging patients potentially requiring replanning using changes in density information between the planning CT and cone-beam CT (CBCT) scan. Potential candidates were defined based on anatomical changes in the primary tumour, mediastinum and a region including the lungs. In total, CBCTs of 70 NSCLC patients were retrospectively scored by two independent experts for the need of replanning using the clinical guidelines. Expert evaluation was used as benchmark for the automatic method and accuracy was quantified using a leave-one-out cross-validation strategy. Results: The automatic method correctly selected 79%, 71%, and 89% of the pre-defined candidates based on anatomical change in the tumour, mediastinum and lungs, respectively. The false positive rate was 27%, 9%, and 19% for these regions, respectively. If a single criterion ('flag') per patient is sufficient for replanning, the overall sensitivity was 89% with a false positive rate of 34%. Most common reasons for false positives were changes in breathing pattern or anatomical shifts outside the evaluated region. Excluding these reasons, the false positive rate dropped to 21% and the accuracy was almost similar to expert evaluation: overall sensitivity of approx. 95% with a false positive rate below 19%. Conclusion: Our method based on daily CBCT allows automatic selection of patients that should be investigated for treatment adaptation.

Anatomic change over the course of treatment for non–small cell lung cancer patients and its impact on intensity-modulated radiation therapy and passive-scattering proton therapy deliveries

“…The action level of ART adoption varies with institutional practice and treatment delivery technique. Van den Bosch et al [32] developed an automatic method to select patients eligible for ART with an accuracy of 79%. The criterion set was the change in the water-equivalent path length to the edge of the target on daily CBCT.…”

Practical Clinical Workflows for Online and Offline Adaptive Radiation Therapy

“…Other means to support assessment and ART decision making are emerging from the groundswell of machine learning and deep learning algorithms entering many fields related to image analysis, including radiotherapy. Such developing approaches produce a model to predict when to adapt, based on learning optimal time points to adapt with supervised data (though, for example, physician determination of whether a fraction should be adapted or not) 60,61 . Deep learning has recently shown very promising results in autosegmentation of organs at risk and even targets, which if clinically deployed could help improve the efficiency of ART.…”

Patient‐specific PTV margins for liver stereotactic body radiation therapy determined using support vector classification with an early warning system for margin adaptation

“…40 The accuracy of the testing set indicates the probability of correct predictions for both classes. In this study, parameter tuning is performed using a grid search to exhaustively loop over all possible combinations of the number of PCs (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20), the C values (0.01, 0.1, 1, 10, and 100), the relative class weights (Class 1: Class 2 = 1:0.8, 1:0.9, and 1:1), and the kernel type (linear, polynomial, and RBF). A fraction of the features is removed using recursive feature elimination.…”

“…Target motion and deformation during treatments should be taken into consideration for treatment planning and delivery. 1,2 Statistical analysis, [3][4][5][6][7] machine learning, [8][9][10][11][12][13][14][15][16][17][18][19] and deep learning 20 using image and posttreatment delivery system log file data are increasingly being reported on for patient-specific and adaptive radiation treatments. However, there are fewer applications of machine learning/deep learning for patient-specific treatments using stereotactic body radiation therapy (SBRT) for liver lesions.…”