Evaluation of deep learning-based deliverable VMAT plan generated by prototype software for automated planning for prostate cancer patients.

This study aims to evaluate the dosimetric accuracy of a deep learning (DL)-based deliverable volumetric arc radiation therapy (VMAT) plan generated using DL-based automated planning assistant system (AIVOT, prototype version) for patients with prostate cancer. The VMAT data (cliDose) of 68 patients with prostate cancer treated with VMAT treatment (70-74 Gy/28-37 fr) at our hospital were used (n = 55 for training and n = 13 for testing). First, a HD-U-net-based 3D dose prediction model implemented in AIVOT was customized using the VMAT data. Thus, a predictive VMAT plan (preDose) comprising AIVOT that predicted the 3D doses was generated. Second, deliverable VMAT plans (deliDose) were created using AIVOT, the radiation treatment planning system Eclipse (version 15.6) and its vender-supplied objective functions. Finally, we compared these two estimated DL-based VMAT treatment plans-i.e. preDose and deliDose-with cliDose. The average absolute dose difference of all DVH parameters for the target tissue between cliDose and deliDose across all patients was 1.32 ± 1.35% (range: 0.04-6.21%), while that for all the organs at risks was 2.08 ± 2.79% (range: 0.00-15.4%). The deliDose was superior to the cliDose in all DVH parameters for bladder and rectum. The blinded plan scoring of deliDose and cliDose was 4.54 ± 0.50 and 5.0 ± 0.0, respectively (All plans scored ≥4 points, P = 0.03.) This study demonstrated that DL-based deliverable plan for prostate cancer achieved the clinically acceptable level. Thus, the AIVOT software exhibited a potential for automated planning with no intervention for patients with prostate cancer.

Improvement of deep learning prediction model in patient-specific QA for VMAT with MLC leaf position map and patient's dose distribution.

PURPOSE: Deep learning-based virtual patient-specific quality assurance (QA) is a novel technique that enables patient QA without measurement. However, this method could be improved by further evaluating the optimal data to be used as input. Therefore, a deep learning-based model that uses multileaf collimator (MLC) information per control point and dose distribution in patient's CT as inputs was developed. METHODS: Overall, 96 volumetric-modulated arc therapy plans generated for prostate cancer treatment were used. We developed a model (Model 1) that can predict measurement-based gamma passing rate (GPR) for a treatment plan using data stored as a map reflecting the MLC leaf position at each control point (MLPM) and data of the dose distribution in patient's CT as inputs. The evaluation of the model was based on the mean absolute error (MAE) and Pearson's correlation coefficient (r) between the measured and predicted GPR. For comparison, we also analyzed models trained with the dose distribution in patient's CT alone (Model 2) and with dose distributions recalculated on a virtual phantom CT (Model 3). RESULTS: At the 2%/2 mm criterion, MAE[%] and r for Model 1, Model 2, and Model 3 were 2.32% ± 0.43% and 0.54 ± 0.03, 2.70% ± 0.26%, and 0.32 ± 0.08, and 2.96% ± 0.23% and 0.24 ± 0.22, respectively; at the 3%/3 mm criterion, these values were 1.25% ± 0.05% and 0.36 ± 0.18, 1.57% ± 0.35% and 0.19 ± 0.20, and 1.39% ± 0.32% and 0.17 ± 0.22, respectively. This result showed that Model 1 exhibited the lowest MAE and highest r at both criteria of 2%/2 mm and 3%3 mm. CONCLUSIONS: These findings showed that a model that combines the MLPM and dose distribution in patient's CT exhibited a better GPR prediction performance compared with the other two studied models.

Assessment of a computed tomography-based radiomics approach for assessing lung function in lung cancer patients.

PURPOSE: We aimed to assess radiomics approaches for estimating three pulmonary function test (PFT) results (forced expiratory volume in one second [FEV1], forced vital capacity [FVC], and the ratio of FEV1 to FVC [FEV1/FVC]) using data extracted from chest computed tomography (CT) images. METHODS: This retrospective study included 85 lung cancer patients (mean age, 75 years ±8; 69 men) who underwent stereotactic body radiotherapy between 2012 and 2020. Their pretreatment chest breath-hold CT and PFT data before radiotherapy were obtained. A total of 107 radiomics features (Shape: 14, Intensity: 18, Texture: 75) were extracted using two methods: extraction of the lung tissue (<-250 HU) (APPROACH 1), and extraction of small blood vessels and lung tissue (APPROACH 2). The PFT results were estimated using the least absolute shrinkage and selection operator regression. Pearson's correlation coefficients (r) were determined for all PFT results, and the area under the curve (AUC) was calculated for FEV1/FVC (<70 %). Finally, we compared our approaches with the conventional formula (Conventional). RESULTS: For the estimated FEV1/FVC, the Pearson's r were 0.21 (P =.06), 0.69 (P <.01), and 0.73 (P <.01) for Conventional, APPROACH 1, and APPROACH 2, respectively; the AUCs for FEV1/FVC (<70 %) were 0.67 (95 % confidence interval [CI]: 0.55, 0.79), 0.82 (CI: 0.72, 0.91; P =.047) and 0.86 (CI: 0.78, 0.94; P =.01), respectively. CONCLUSIONS: The radiomics approach performed better than the conventional equation and may be useful for assessing lung function based on CT images.

Feasibility of Differential Dose-Volume Histogram Features in Multivariate Prediction Model for Radiation Pneumonitis Occurrence.

The purpose of this study is to introduce differential dose−volume histogram (dDVH) features into machine learning for radiation pneumonitis (RP) prediction and to demonstrate the predictive performance of the developed model based on integrated cumulative dose−volume histogram (cDVH) and dDVH features. Materials and methods: cDVH and dDVH features were calculated for 153 patients treated for non-small-cell lung cancer with 60−66 Gy and dose bins ranging from 2 to 8 Gy in 2 Gy increments. RP prediction models were developed with the least absolute shrinkage and selection operator (LASSO) through fivefold cross-validation. Results: Among the 152 patients in the patient cohort, 41 presented ≥grade 2 RP. The interdependencies between cDVH features evaluated by Spearman's correlation were significantly resolved by the inclusion of dDVH features. The average area under curve for the RP prediction model using cDVH and dDVH model was 0.73, which was higher than the average area under curve using cDVH model for 0.62 with statistically significance (p < 0.01). An analysis using the entire set of regression coefficients determined by LASSO demonstrated that dDVH features represented four of the top five frequently selected features in the model fitting, regardless of dose bin. Conclusions: We successfully developed an RP prediction model that integrated cDVH and dDVH features. The best RP prediction model was achieved using dDVH (dose bin = 4 Gy) features in the machine learning process.

A deep learning-based radiomics approach to predict head and neck tumor regression for adaptive radiotherapy.

Early regression-the regression in tumor volume during the initial phase of radiotherapy (approximately 2 weeks after treatment initiation)-is a common occurrence during radiotherapy. This rapid radiation-induced tumor regression may alter target coordinates, necessitating adaptive radiotherapy (ART). We developed a deep learning-based radiomics (DLR) approach to predict early head and neck tumor regression and thereby facilitate ART. Primary gross tumor volume (GTVp) was monitored in 96 patients and nodal GTV (GTVn) in 79 patients during treatment. All patients underwent two computed tomography (CT) scans: one before the start of radiotherapy for initial planning and one during radiotherapy for boost planning. Patients were assigned to regression and nonregression groups according to their median tumor regression rate (ΔGTV/treatment day from initial to boost CT scan). We input a GTV image into the convolutional neural network model, which was pretrained using natural image datasets, via transfer learning. The deep features were extracted from the last fully connected layer. To clarify the prognostic power of the deep features, machine learning models were trained. The models then predicted the regression and nonregression of GTVp and GTVn and evaluated the predictive performance by 0.632 + bootstrap area under the curve (AUC). Predictive performance for GTVp regression was highest using the InceptionResNetv2 model (mean AUC = 0.75) and that for GTVn was highest using NASNetLarge (mean AUC = 0.73). Both models outperformed the handcrafted radiomics features (mean AUC = 0.63 for GTVp and 0.61 for GTVn) or clinical factors (0.64 and 0.67, respectively). DLR may facilitate ART for improved radiation side-effects and target coverage.

Impact of feature selection methods and subgroup factors on prognostic analysis with CT-based radiomics in non-small cell lung cancer patients.

BACKGROUND: Radiomics is a new technology to noninvasively predict survival prognosis with quantitative features extracted from medical images. Most radiomics-based prognostic studies of non-small-cell lung cancer (NSCLC) patients have used mixed datasets of different subgroups. Therefore, we investigated the radiomics-based survival prediction of NSCLC patients by focusing on subgroups with identical characteristics. METHODS: A total of 304 NSCLC (Stages I-IV) patients treated with radiotherapy in our hospital were used. We extracted 107 radiomic features (i.e., 14 shape features, 18 first-order statistical features, and 75 texture features) from the gross tumor volume drawn on the free breathing planning computed tomography image. Three feature selection methods [i.e., test-retest and multiple segmentation (FS1), Pearson's correlation analysis (FS2), and a method that combined FS1 and FS2 (FS3)] were used to clarify how they affect survival prediction performance. Subgroup analysis for each histological subtype and each T stage applied the best selection method for the analysis of All data. We used a least absolute shrinkage and selection operator Cox regression model for all analyses and evaluated prognostic performance using the concordance-index (C-index) and the Kaplan-Meier method. For subgroup analysis, fivefold cross-validation was applied to ensure model reliability. RESULTS: In the analysis of All data, the C-index for the test dataset is 0.62 (FS1), 0.63 (FS2), and 0.62 (FS3). The subgroup analysis indicated that the prediction model based on specific histological subtypes and T stages had a higher C-index for the test dataset than that based on All data (All data, 0.64 vs. SCCall, 060; ADCall, 0.69; T1, 0.68; T2, 0.65; T3, 0.66; T4, 0.70). In addition, the prediction models unified for each T stage in histological subtype showed a different trend in the C-index for the test dataset between ADC-related and SCC-related models (ADCT1-ADCT4, 0.72-0.83; SCCT1-SCCT4, 0.58-0.71). CONCLUSIONS: Our results showed that feature selection methods moderately affected the survival prediction performance. In addition, prediction models based on specific subgroups may improve the prediction performance. These results may prove useful for determining the optimal radiomics-based predication model.