Combination of Clinical Factors and Radiomics Can Predict Local Recurrence and Metastasis After Stereotactic Body Radiotherapy for Non-small Cell Lung Cancer.

BACKGROUND/AIM: Radiomics, which links radiological image features with patient prognoses, is expected to be applied for the prediction of the clinical outcomes of radiotherapy. We investigated the clinical and radiomic factors associated with recurrence patterns after stereotactic body radiotherapy (SBRT) for non-small cell lung cancer (NSCLC). PATIENTS AND METHODS: We retrospectively analyzed 125 patients with histologically confirmed NSCLC who underwent SBRT between April 2003 and June 2017 at our institution. A radiomic score was calculated from five radiomics features (histogram and texture features) selected using the LASSO Cox regression model. These features were extracted from the gross tumor volume (GTV) in three-dimensional wavelet decomposition CT images. We used univariate and multivariate analyses to determine the associations between local control (LC) time and metastasis-free survival (MFS), clinical factors (age, sex, performance status, operability, smoking, histology, and tumor diameter), and the radiomic score. RESULTS: With a median follow-up of 37 months, the following 3-year rates were observed: overall survival, 80.9%; progression-free survival, 61.7%; LC, 75.1%, and MFS; 74.5%. In multivariate analysis, histology (squamous cell carcinoma vs. non-squamous cell carcinoma, p=0.0045), tumor diameter (>3 cm vs. ≤3 cm, p=0.039); and radiomic score (>0.043 vs. ≤0.043, p=0.042) were significantly associated with LC, and the radiomic score (>0.304 vs. ≤0.304, p<0.001) was significantly associated with MFS. CONCLUSION: Histology, tumor diameter, and radiomic score could be significant factors for predicting NSCLC recurrence patterns after SBRT.

Magnetic resonance-based imaging biopsy with signatures including topological Betti number features for prediction of primary brain metastatic sites.

This study incorporated topology Betti number (BN) features into the prediction of primary sites of brain metastases and the construction of magnetic resonance-based imaging biopsy (MRB) models. The significant features of the MRB model were selected from those obtained from gray-scale and three-dimensional wavelet-filtered images, BN and inverted BN (iBN) maps, and clinical variables (age and gender). The primary sites were predicted as either lung cancer or other cancers using MRB models, which were built using seven machine learning methods with significant features chosen by three feature selection methods followed by a combination strategy. Our study dealt with a dataset with relatively smaller brain metastases, which included effective diameters greater than 2 mm, with metastases ranging from 2 to 9 mm accounting for 17% of the dataset. The MRB models were trained by T1-weighted contrast-enhanced images of 494 metastases chosen from 247 patients and applied to 115 metastases from 62 test patients. The most feasible model attained an area under the receiver operating characteristic curve (AUC) of 0.763 for the test patients when using a signature including features of BN and iBN maps, gray-scale and wavelet-filtered images, and clinical variables. The AUCs of the model were 0.744 for non-small cell lung cancer and 0.861 for small cell lung cancer. The results suggest that the BN signature boosted the performance of MRB for the identification of primary sites of brain metastases including small tumors.

Three-dimensional topological radiogenomics of epidermal growth factor receptor Del19 and L858R mutation subtypes on computed tomography images of lung cancer patients.

OBJECTIVES: To elucidate a novel radiogenomics approach using three-dimensional (3D) topologically invariant Betti numbers (BNs) for topological characterization of epidermal growth factor receptor (EGFR) Del19 and L858R mutation subtypes. METHODS: In total, 154 patients (wild-type EGFR, 72 patients; Del19 mutation, 45 patients; and L858R mutation, 37 patients) were retrospectively enrolled and randomly divided into 92 training and 62 test cases. Two support vector machine (SVM) models to distinguish between wild-type and mutant EGFR (mutation [M] classification) as well as between the Del19 and L858R subtypes (subtype [S] classification) were trained using 3DBN features. These features were computed from 3DBN maps by using histogram and texture analyses. The 3DBN maps were generated using computed tomography (CT) images based on the Cech complex constructed on sets of points in the images. These points were defined by coordinates of voxels with CT values higher than several threshold values. The M classification model was built using image features and demographic parameters of sex and smoking status. The SVM models were evaluated by determining their classification accuracies. The feasibility of the 3DBN model was compared with those of conventional radiomic models based on pseudo-3D BN (p3DBN), two-dimensional BN (2DBN), and CT and wavelet-decomposition (WD) images. The validation of the model was repeated with 100 times random sampling. RESULTS: The mean test accuracies for M classification with 3DBN, p3DBN, 2DBN, CT, and WD images were 0.810, 0.733, 0.838, 0.782, and 0.799, respectively. The mean test accuracies for S classification with 3DBN, p3DBN, 2DBN, CT, and WD images were 0.773, 0.694, 0.657, 0.581, and 0.696, respectively. CONCLUSION: 3DBN features, which showed a radiogenomic association with the characteristics of the EGFR Del19/L858R mutation subtypes, yielded higher accuracy for subtype classifications in comparison with conventional features.

Topology-based radiomic features for prediction of parotid gland cancer malignancy grade in magnetic resonance images.

PURPOSE: The malignancy grades of parotid gland cancer (PGC) have been assessed for a decision of treatment policies. Therefore, we have investigated the feasibility of topology-based radiomic features for the prediction of parotid gland cancer (PGC) malignancy grade in magnetic resonance (MR) images. MATERIALS AND METHODS: Two-dimensional T1- and T2-weighted MR images of 39 patients with PGC were selected for this study. Imaging properties of PGC can be quantified using the topology, which could be useful for assessing the number of the k-dimensional holes or heterogeneity in PGC regions using invariants of the Betti numbers. Radiomic signatures were constructed from 41,472 features obtained after a harmonization using an elastic net model. PGC patients were stratified using a logistic classification into low/intermediate- and high-grade malignancy groups. The training data were increased by four times to avoid the overfitting problem using a synthetic minority oversampling technique. The proposed approach was assessed using a 4-fold cross-validation test. RESULTS: The highest accuracy of the proposed approach was 0.975 for the validation cases, whereas that of the conventional approach was 0.694. CONCLUSION: This study indicated that topology-based radiomic features could be feasible for the noninvasive prediction of the malignancy grade of PGCs.

Dual segmentation models for poorly and well-differentiated hepatocellular carcinoma using two-step transfer deep learning on dynamic contrast-enhanced CT images.

The aim of this study was to develop dual segmentation models for poorly and well-differentiated hepatocellular carcinoma (HCC), using two-step transfer learning (TSTL) based on dynamic contrast-enhanced (DCE) computed tomography (CT) images. From 2013 to 2019, DCE-CT images of 128 patients with 80 poorly differentiated and 48 well-differentiated HCCs were selected at our hospital. In the first transfer learning (TL) step, a pre-trained segmentation model with 192 CT images of lung cancer patients was retrained as a poorly differentiated HCC model. In the second TL step, a well-differentiated HCC model was built from a poorly differentiated HCC model. The average three-dimensional Dice's similarity coefficient (3D-DSC) and 95th-percentile of the Hausdorff distance (95% HD) were mainly employed to evaluate the segmentation accuracy, based on a nested fourfold cross-validation test. The DSC denotes the degree of regional similarity between the HCC reference regions and the regions estimated using the proposed models. The 95% HD is defined as the 95th-percentile of the maximum measures of how far two subsets of a metric space are from each other. The average 3D-DSC and 95% HD were 0.849 ± 0.078 and 1.98 ± 0.71 mm, respectively, for poorly differentiated HCC regions, and 0.811 ± 0.089 and 2.01 ± 0.84 mm, respectively, for well-differentiated HCC regions. The average 3D-DSC for both regions was 1.2 times superior to that calculated without the TSTL. The proposed model using TSTL from the lung cancer dataset showed the potential to segment poorly and well-differentiated HCC regions on DCE-CT images.

Can Persistent Homology Features Capture More Intrinsic Information about Tumors from 18F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography Images of Head and Neck Cancer Patients?

This study hypothesized that persistent homology (PH) features could capture more intrinsic information about the metabolism and morphology of tumors from 18F-fluorodeoxyglucose positron emission tomography (PET)/computed tomography (CT) images of patients with head and neck (HN) cancer than other conventional features. PET/CT images and clinical variables of 207 patients were selected from the publicly available dataset of the Cancer Imaging Archive. PH images were generated from persistent diagrams obtained from PET/CT images. The PH features were derived from the PH PET/CT images. The signatures were constructed in a training cohort from features from CT, PET, PH-CT, and PH-PET images; clinical variables; and the combination of features and clinical variables. Signatures were evaluated using statistically significant differences (p-value, log-rank test) between survival curves for low- and high-risk groups and the C-index. In an independent test cohort, the signature consisting of PH-PET features and clinical variables exhibited the lowest log-rank p-value of 3.30 × 10-5 and C-index of 0.80, compared with log-rank p-values from 3.52 × 10-2 to 1.15 × 10-4 and C-indices from 0.34 to 0.79 for other signatures. This result suggests that PH features can capture the intrinsic information of tumors and predict prognosis in patients with HN cancer.

Relapse predictability of topological signature on pretreatment planning CT images of stage I non-small cell lung cancer patients before treatment with stereotactic ablative radiotherapy.

BACKGROUND: This study aimed to explore the predictability of topological signatures linked to the locoregional relapse (LRR) and distant metastasis (DM) on pretreatment planning computed tomography images of stage I non-small cell lung cancer (NSCLC) patients before treatment with stereotactic ablative radiotherapy (SABR). METHODS: We divided 125 primary stage I NSCLC patients (LRR: 34, DM: 22) into training (n = 60) and test datasets (n = 65), and the training dataset was augmented to 260 cases using a synthetic minority oversampling technique. The relapse predictabilities of the conventional wavelet-based features (WF), topology-based features [BF, Betti number (BN) map features; iBF, inverted BN map features], and their combined features (BWF, iBWF) were compared. The patients were stratified into high-risk and low-risk groups using the medians of the radiomics scores in the training dataset. RESULTS: For the LRR in the test, the iBF, iBWF, and WF showed statistically significant differences (p < 0.05), and the highest nLPC was obtained for the iBF. For the DM in the test, the iBWF showed a significant difference and the highest nLPC. CONCLUSION: The iBF indicated the potential of improving the LRR and DM prediction of stage I NSCLC patients prior to undergoing SABR.

Synergistic combination of a topologically invariant imaging signature and a biomarker for the accurate prediction of symptomatic radiation pneumonitis before stereotactic ablative radiotherapy for lung cancer: A retrospective analysis.

OBJECTIVES: We aimed to explore the synergistic combination of a topologically invariant Betti number (BN)-based signature and a biomarker for the accurate prediction of symptomatic (grade ≥2) radiation-induced pneumonitis (RP+) before stereotactic ablative radiotherapy (SABR) for lung cancer. METHODS: A total of 272 SABR cases with early-stage non-small cell lung cancer were chosen for this study. The occurrence of RP+ was predicted using a support vector machine (SVM) model trained with the combined features of the BN-based signature extracted from planning computed tomography (pCT) images and a pretreatment biomarker, serum Krebs von den Lungen-6 (BN+KL-6 model). In all, 242 (20 RP+ and 222 RP-(grade 1)) and 30 cases (8 RP+ and 22 RP-) were used for training and testing the model, respectively. The BN-based features were extracted from BN maps that characterize topologically invariant heterogeneous traits of potential RP+ lung regions on pCT images by applying histogram- and texture-based feature calculations to the maps. The SVM models were built to predict RP+ patients with a BN signature that was constructed based on the least absolute shrinkage and selection operator logistic regression model. The evaluation of the prediction models was performed based on the area under the receiver operating characteristic curves (AUCs) and accuracy in the test. The performance of the BN+KL-6 model was compared to the performance based on the BN, conventional original pCT, and wavelet decomposition (WD) models. RESULTS: The test AUCs obtained for the BN+KL-6, BN, pCT, and WD models were 0.825, 0.807, 0.642, and 0.545, respectively. The accuracies of the BN+KL-6, BN, pCT, and WD models were found to be 0.724, 0.708, 0.591, and 0.534, respectively. CONCLUSION: This study demonstrated the comprehensive performance of the BN+KL-6 model for the prediction of potential RP+ patients before SABR for lung cancer.

Stratification of prostate cancer patients into low- and high-grade groups using multiparametric magnetic resonance radiomics with dynamic contrast-enhanced image joint histograms.

PURPOSE: This study aimed to investigate the potential of stratification of prostate cancer patients into low- and high-grade groups (GGs) using multiparametric magnetic resonance (mpMR) radiomics in conjunction with two-dimensional (2D) joint histograms computed with dynamic contrast-enhanced (DCE) images. METHODS: A total of 101 prostate cancer regions extracted from the MR images of 44 patients were identified and divided into training (n = 31 with 72 cancer regions) and test datasets (n = 13 with 29 cancer regions). Each dataset included low-grade tumors (International Society of Urological Pathology [ISUP] GG ≤ 2) and high-grade tumors (ISUP GG ≥ 3). A total of 137,970 features consisted of mpMR image (16 types of images in four sequences)-based and joint histogram (DCE images at 10 phases)-based features for each cancer region. Joint histogram features can visualize temporally changing perfusion patterns in prostate cancer based on the joint histograms between different phases or subtraction phases of DCE images. Nine signatures (a set of significant features related to GGs) were determined using the best combinations of features selected using the least absolute shrinkage and selection operator. Further, support vector machine models with the nine signatures were built based on a leave-one-out cross-validation for the training dataset and evaluated with receiver operating characteristic (ROC) curve analysis. RESULTS: The signature showing the best performance was constructed using six features derived from the joint histograms, DCE original images, and apparent diffusion coefficient maps. The areas under the ROC curves for the training and test datasets were 1.00 and 0.985, respectively. CONCLUSION: This study suggests that the proposed approach with mpMR radiomics in conjunction with 2D joint histogram computed with DCE images could have the potential to stratify prostate cancer patients into low- and high-GGs.

Robust radiogenomics approach to the identification of EGFR mutations among patients with NSCLC from three different countries using topologically invariant Betti numbers.

OBJECTIVES: To propose a novel robust radiogenomics approach to the identification of epidermal growth factor receptor (EGFR) mutations among patients with non-small cell lung cancer (NSCLC) using Betti numbers (BNs). MATERIALS AND METHODS: Contrast enhanced computed tomography (CT) images of 194 multi-racial NSCLC patients (79 EGFR mutants and 115 wildtypes) were collected from three different countries using 5 manufacturers' scanners with a variety of scanning parameters. Ninety-nine cases obtained from the University of Malaya Medical Centre (UMMC) in Malaysia were used for training and validation procedures. Forty-one cases collected from the Kyushu University Hospital (KUH) in Japan and fifty-four cases obtained from The Cancer Imaging Archive (TCIA) in America were used for a test procedure. Radiomic features were obtained from BN maps, which represent topologically invariant heterogeneous characteristics of lung cancer on CT images, by applying histogram- and texture-based feature computations. A BN-based signature was determined using support vector machine (SVM) models with the best combination of features that maximized a robustness index (RI) which defined a higher total area under receiver operating characteristics curves (AUCs) and lower difference of AUCs between the training and the validation. The SVM model was built using the signature and optimized in a five-fold cross validation. The BN-based model was compared to conventional original image (OI)- and wavelet-decomposition (WD)-based models with respect to the RI between the validation and the test. RESULTS: The BN-based model showed a higher RI of 1.51 compared with the models based on the OI (RI: 1.33) and the WD (RI: 1.29). CONCLUSION: The proposed model showed higher robustness than the conventional models in the identification of EGFR mutations among NSCLC patients. The results suggested the robustness of the BN-based approach against variations in image scanner/scanning parameters.

Radiomic features based on Hessian index for prediction of prognosis in head-and-neck cancer patients.

This study demonstrated the usefulness of radiomic features based on the Hessian index of differential topology for the prediction of prognosis prior to treatment in head-and-neck (HN) cancer patients. The Hessian index, which can indicate tumor heterogeneity with convex, concave, and other points (saddle points), was calculated as the number of negative eigenvalues of the Hessian matrix at each voxel on computed tomography (CT) images. Three types of signatures were constructed in a training cohort (n = 126), one type each from CT conventional features, Hessian index features, and combined features from the conventional and index feature sets. The prognostic value of the signatures were evaluated using statistically significant difference (p value, log-rank test) to compare the survival curves of low- and high-risk groups. In a test cohort (n = 68), the p values of the models built with conventional, index, combined features, and clinical variables were 2.95 [Formula: see text] 10-2, 1.85 [Formula: see text] 10-2, 3.17 [Formula: see text] 10-2, and 1.87 [Formula: see text] 10-3, respectively. When the features were integrated with clinical variables, the p values of conventional, index, and combined features were 3.53 [Formula: see text] 10-3, 1.28 [Formula: see text] 10-3, and 1.45 [Formula: see text] 10-3, respectively. This result indicates that index features could provide more prognostic information than conventional features and further increase the prognostic value of clinical variables in HN cancer patients.

Radiomic prediction of radiation pneumonitis on pretreatment planning computed tomography images prior to lung cancer stereotactic body radiation therapy.

This study developed a radiomics-based predictive model for radiation-induced pneumonitis (RP) after lung cancer stereotactic body radiation therapy (SBRT) on pretreatment planning computed tomography (CT) images. For the RP prediction models, 275 non-small-cell lung cancer patients consisted of 245 training (22 with grade ≥ 2 RP) and 30 test cases (8 with grade ≥ 2 RP) were selected. A total of 486 radiomic features were calculated to quantify the RP texture patterns reflecting radiation-induced tissue reaction within lung volumes irradiated with more than x Gy, which were defined as LVx. Ten subsets consisting of all 22 RP cases and 22 or 23 randomly selected non-RP cases were created from the imbalanced dataset of 245 training patients. For each subset, signatures were constructed, and predictive models were built using the least absolute shrinkage and selection operator logistic regression. An ensemble averaging model was built by averaging the RP probabilities of the 10 models. The best model areas under the receiver operating characteristic curves (AUCs) calculated on the training and test cohort for LV5 were 0.871 and 0.756, respectively. The radiomic features calculated on pretreatment planning CT images could be predictive imaging biomarkers for RP after lung cancer SBRT.

Semi-automated prediction approach of target shifts using machine learning with anatomical features between planning and pretreatment CT images in prostate radiotherapy.

The goal of this study was to develop a semi-automated prediction approach of target shifts using machine learning architecture (MLA) with anatomical features for prostate radiotherapy. Our hypothesis was that anatomical features between planning computed tomography (pCT) and pretreatment cone-beam computed tomography (CBCT) images could be used to predict the target, i.e. clinical target volume (CTV) shifts, with small errors. The pCT and daily CBCT images of 20 patients with prostate cancer were selected. The first 10 patients were employed for the development, and the second 10 patients for a validation test. The CTV position errors between the pCT and CBCT images were determined as reference CTV shifts (teacher data) after an automated bone-based registration. The anatomical features associated with rectum, bladder and prostate were calculated from the pCT and CBCT images. The features were fed as the input with the teacher data into five MLAs, i.e. three types of artificial neural networks, support vector regression (SVR) and random forests. Since the CTV shifts along the left-right direction were negligible, the MLAs were developed along the superior-inferior and anterior-posterior directions. The proposed framework was evaluated from the residual errors between the reference and predicted CTV shifts. In the validation test, the mean residual error with its standard deviation was 1.01 ± 1.09 mm in SVR using only one feature (one click), which was associated with positional difference of the upper rectal wall. The results suggested that MLAs with anatomical features could be useful in prediction of CTV shifts for prostate radiotherapy.

Automated approach for estimation of grade groups for prostate cancer based on histological image feature analysis.

BACKGROUND: There is a low reproducibility of the Gleason scores that determine the grade group of prostate cancer given the intra- and interobserver variability among pathologists. This study aimed to develop an automated approach for estimating prostate cancer grade groups based on features obtained from histological image analysis. METHODS: Fifty-nine patients who underwent radical prostatectomy were selected under the approval of the institutional review board of our university hospital. For estimation, we followed the grade group criteria provided by the International Society of Urological Pathology in 2014. One hundred eight specimen slides obtained from the patients were digitized to extract 110 regions of interest (ROI) from hematoxylin and eosin-stained histological images using a digital whole slide scanner at ×20 magnification with a pixel size of 0.4 µm. Each color pixel value in the ROI was decomposed into six intensities corresponding to the RGB (red, green, and blue) and HSV (hue, saturation, and value) color models. Image features were extracted by histological image analysis, obtaining 54 features from the ROI based on histogram and texture analyses in the six types of decomposed histological images. Then, 40 representative features were selected from the 324 histological image features based on statistically significant differences (P < .05) between the mean image feature values for high (≥3, Gleason score ≥4 + 3) and low (≤2, Gleason score ≤3 + 4) grade groups. The relationship between grade groups and the most representative image feature (ie, complexity) was approximated using regression to estimate real-number grade groups defined by continuous numerical grading. Finally, the grade groups were expressed as the conventional grade groups (ie, integers from 1 to 5) using a piecewise step function. RESULTS: The grade groups were correctly estimated by the proposed approach without errors on training (70 ROIs) and validation (40 ROIs) data. CONCLUSIONS: Our results suggest that the proposed approach may support pathologists during the evaluation of grade groups for prostate cancer, thus mitigating intra- and interobserver variability.

Homological radiomics analysis for prognostic prediction in lung cancer patients.

PURPOSE: This study explored a novel homological analysis method for prognostic prediction in lung cancer patients. MATERIALS AND METHODS: The potential of homology-based radiomic features (HFs) was investigated by comparing HFs to conventional wavelet-based radiomic features (WFs) and combined radiomic features consisting of HFs and WFs (HWFs), using training (n = 135) and validation (n = 70) datasets, and Kaplan-Meier analysis. A total of 13,824 HFs were derived through homology-based texture analysis using Betti numbers, which represent the topologically invariant morphological characteristics of lung cancer. The prognostic potential of HFs was evaluated using statistically significant differences (p-values, log-rank test) to compare the survival curves of high- and low-risk patients. Those patients were stratified into high- and low-risk groups using the medians of the radiomic scores of signatures constructed with an elastic-net-regularized Cox proportional hazard model. Furthermore, deep learning (DL) based on AlexNet was utilized to compare HFs by stratifying patients into the two groups using a network that was pre-trained with over one million natural images from an ImageNet database. RESULTS: For the training dataset, the p-values between the two survival curves were 6.7 × 10-6 (HF), 5.9 × 10-3 (WF), 7.4 × 10-6 (HWF), and 1.1 × 10-3 (DL). The p-values for the validation dataset were 3.4 × 10-5 (HF), 6.7 × 10-1 (WF), 1.7 × 10-7 (HWF), and 1.2 × 10-1 (DL). CONCLUSION: This study demonstrates the excellent potential of HFs for prognostic prediction in lung cancer patients.

Radiomics with artificial intelligence for precision medicine in radiation therapy.

Recently, the concept of radiomics has emerged from radiation oncology. It is a novel approach for solving the issues of precision medicine and how it can be performed, based on multimodality medical images that are non-invasive, fast and low in cost. Radiomics is the comprehensive analysis of massive numbers of medical images in order to extract a large number of phenotypic features (radiomic biomarkers) reflecting cancer traits, and it explores the associations between the features and patients' prognoses in order to improve decision-making in precision medicine. Individual patients can be stratified into subtypes based on radiomic biomarkers that contain information about cancer traits that determine the patient's prognosis. Machine-learning algorithms of AI are boosting the powers of radiomics for prediction of prognoses or factors associated with treatment strategies, such as survival time, recurrence, adverse events, and subtypes. Therefore, radiomic approaches, in combination with AI, may potentially enable practical use of precision medicine in radiation therapy by predicting outcomes and toxicity for individual patients.

Potentials of radiomics for cancer diagnosis and treatment in comparison with computer-aided diagnosis.

Computer-aided diagnosis (CAD) is a field that is essentially based on pattern recognition that improves the accuracy of a diagnosis made by a physician who takes into account the computer's "opinion" derived from the quantitative analysis of radiological images. Radiomics is a field based on data science that massively and comprehensively analyzes a large number of medical images to extract a large number of phenotypic features reflecting disease traits, and explores the associations between the features and patients' prognoses for precision medicine. According to the definitions for both, you may think that radiomics is not a paraphrase of CAD, but you may also think that these definitions are "image manipulation". However, there are common and different features between the two fields. This review paper elaborates on these common and different features and introduces the potential of radiomics for cancer diagnosis and treatment by comparing it with CAD.

Feasibility of anatomical feature points for the estimation of prostate locations in the Bayesian delineation frameworks for prostate cancer radiotherapy.

This study aimed to investigate the feasibility of anatomical feature points for the estimation of prostate locations in the Bayesian delineation frameworks for prostate cancer radiotherapy. The relationships between the reference centroids of prostate regions (CPRs) (prostate locations) and anatomical feature points were explored, and the most feasible anatomical feature points were selected based on the smallest location estimation errors of CPRs and the largest Dice's similarity coefficient (DSC) between the reference and extracted prostates. The reference CPRs were calculated according to reference prostate contours determined by radiation oncologists. Five anatomical feature points were manually determined on a prostate, bladder, and rectum in a sagittal plane of a planning computed tomography image for each case. The CPRs were estimated using three machine learning architectures [artificial neural network, random forest, and support vector machine (SVM)], which learned the relationships between the reference CPRs and anatomical feature points. The CPRs were applied for placing a prostate probabilistic atlas at the coordinates and extracting prostate regions using a Bayesian delineation framework. The average estimation errors without and with SVM using three feature points, which indicated the best performance, were 5.6 ± 3.7 mm and 1.8 ± 1.0 mm, respectively (the smallest error) (p < 0.001). The average DSCs without and with SVM using the three feature points were 0.69 ± 0.13 and 0.82 ± 0.055, respectively (the highest DSC) (p < 0.001). The anatomical feature points may be feasible for the estimation of prostate locations, which can be applied to the general Bayesian delineation frameworks for prostate cancer radiotherapy.