Radiomic Feature Robustness and Reproducibility in Quantitative Bone Radiography: A Study on Radiologic Parameter Changes

Magnetic Resonance Image Radiomic Reproducibility: The Impact of Preprocessing on Extracted Features from Gross and High-Risk Clinical Tumor Volumes in Cervical Cancer Patients before Brachytherapy

Background

Radiomic feature reproducibility assessment is critical in radiomics-based image biomarker discovery. This study aims to evaluate the impact of preprocessing parameters on the reproducibility of magnetic resonance image (MRI) radiomic features extracted from gross tumor volume (GTV) and high-risk clinical tumor volume (HR-CTV) in cervical cancer (CC) patients.

Methods

This study included 99 patients with pathologically confirmed cervical cancer who underwent an MRI prior to receiving brachytherapy. The GTV and HR-CTV were delineated on T2-weighted MRI and inputted into 3D Slicer for radiomic analysis. Before feature extraction, all images were preprocessed to a combination of several parameters of Laplacian of Gaussian (1 and 2), resampling (0.5 and 1), and bin width (5, 10, 25, and 50). The reproducibility of radiomic features was analyzed using the intra-class correlation coefficient (ICC).

Results

Almost all shapes and first-order features had ICC values > 0.95. Most second-order texture features were not reproducible (ICC < 0.95) in GTV and HR-CTV. Furthermore, 20% of all neighboring gray-tone difference matrix texture features had ICC > 0.90 in both GTV and HR-CTV.

Conclusion

The results presented here showed that MRI radiomic features are vulnerable to changes in preprocessing, and this issue must be understood and applied before any clinical decision-making. Features with ICC > 0.90 were considered the most reproducible features. Shape and first-order radiomic features were the most reproducible features in both GTV and HR-CTV. Our results also showed that GTV and HR-CTV radiomic features had similar changes against preprocessing sets.

The Applications of Artificial Intelligence in Digestive System Neoplasms: A Review

Importance

Digestive system neoplasms (DSNs) are the leading cause of cancer-related mortality with a 5-year survival rate of less than 20%. Subjective evaluation of medical images including endoscopic images, whole slide images, computed tomography images, and magnetic resonance images plays a vital role in the clinical practice of DSNs, but with limited performance and increased workload of radiologists or pathologists. The application of artificial intelligence (AI) in medical image analysis holds promise to augment the visual interpretation of medical images, which could not only automate the complicated evaluation process but also convert medical images into quantitative imaging features that associated with tumor heterogeneity.

Highlights

We briefly introduce the methodology of AI for medical image analysis and then review its clinical applications including clinical auxiliary diagnosis, assessment of treatment response, and prognosis prediction on 4 typical DSNs including esophageal cancer, gastric cancer, colorectal cancer, and hepatocellular carcinoma.

Conclusion

AI technology has great potential in supporting the clinical diagnosis and treatment decision-making of DSNs. Several technical issues should be overcome before its application into clinical practice of DSNs.

Data harmonisation for information fusion in digital healthcare: A state-of-the-art systematic review, meta-analysis and future research directions

Removing the bias and variance of multicentre data has always been a challenge in large scale digital healthcare studies, which requires the ability to integrate clinical features extracted from data acquired by different scanners and protocols to improve stability and robustness. Previous studies have described various computational approaches to fuse single modality multicentre datasets. However, these surveys rarely focused on evaluation metrics and lacked a checklist for computational data harmonisation studies. In this systematic review, we summarise the computational data harmonisation approaches for multi-modality data in the digital healthcare field, including harmonisation strategies and evaluation metrics based on different theories. In addition, a comprehensive checklist that summarises common practices for data harmonisation studies is proposed to guide researchers to report their research findings more effectively. Last but not least, flowcharts presenting possible ways for methodology and metric selection are proposed and the limitations of different methods have been surveyed for future research.

Using radiomic features of lumbar spine CT images to differentiate osteoporosis from normal bone density

OBJECTIVE

This study aimed to develop a predictive model to detect osteoporosis using radiomic features from lumbar spine computed tomography (CT) images.

METHODS

A total of 133 patients were included in this retrospective study, 41 men and 92 women, with a mean age of 65.45 ± 9.82 years (range: 31-94 years); 53 had normal bone mineral density, 32 osteopenia, and 48 osteoporosis. For each patient, the L1-L4 vertebrae on the CT images were automatically segmented using SenseCare and defined as regions of interest (ROIs). In total, 1,197 radiomic features were extracted from these ROIs using PyRadiomics. The most significant features were selected using logistic regression and Pearson correlation coefficient matrices. Using these features, we constructed three linear classification models based on the random forest (RF), support vector machine (SVM), and K-nearest neighbor (KNN) algorithms, respectively. The training and test sets were repeatedly selected using fivefold cross-validation. The model performance was evaluated using the area under the receiver operator characteristic curve (AUC) and confusion matrix.

RESULTS

The classification model based on RF had the highest performance, with an AUC of 0.994 (95% confidence interval [CI]: 0.979-1.00) for differentiating normal BMD and osteoporosis, 0.866 (95% CI: 0.779-0.954) for osteopenia versus osteoporosis, and 0.940 (95% CI: 0.891-0.989) for normal BMD versus osteopenia.

CONCLUSIONS

The excellent performance of this radiomic model indicates that lumbar spine CT images can effectively be used to identify osteoporosis and as a tool for opportunistic osteoporosis screening.

Bone radiomics reproducibility: a three-centered study on the impacts of image contrast, edge enhancement, and latitude variations

This study aims to measure the reproducibility of radiomics features in ankle bone radiography over changes in post-processing parameters including contrast, edge enhancement and latitude. Lateral ankle bone radiographies for sixty patients were obtained from three digital radiology centers. All images were acquired by same image acquisition settings. A two-dimensional region of interest was drawn in any image and 93 features from 6 feature sets including first and second order were extracted. The coefficient of variation (COV) and intraclass correlation coefficient (ICC) were calculated to assess feature reproducibility for each center and among all centers in three scenarios: Adams (Nat Rev Endocrinol 9(1):28, 2013) ten different contrast Brown et al. (J Med Imaging 5(1):011017, 2018) ten different edge enhancement and Hirvasniemi et al. (Osteoarthr Cartilage 27(6):906-914, 2019)  ten different image latitude parameters. Based on ICC analysis, it is observed that 46-100-44% of Histogram, 54-72-42% of GLCM, 43-76-36% of GLDM, 60-90-17% of GLRLM, 33-19-21% of GLSZM and 13-20-0% of NGTDM radiomics features had 90% < ICC < 100% over changes in contrast-edge enhancement-latitude changes respectively. Based on COV, GLRLM was only feature set that 100% of their features had COV ≤ 5% over changes in contrast and edge enhancement. The results presented here, indicating that radiomics features extracted are vulnerable over changes in contrast, edge enhancement and latitude. The most reproducible features that introduced in this study could be used for further clinical decision making.

Radiomic modeling to predict risk of vertebral compression fracture after stereotactic body radiation therapy for spinal metastases

OBJECTIVE

In the treatment of spinal metastases with stereotactic body radiation therapy (SBRT), vertebral compression fracture (VCF) is a common and potentially morbid complication. Better methods to identify patients at high risk of radiation-induced VCF are needed to evaluate prophylactic measures. Radiomic features from pretreatment imaging may be employed to more accurately predict VCF. The objective of this study was to develop and evaluate a machine learning model based on clinical characteristics and radiomic features from pretreatment imaging to predict the risk of VCF after SBRT for spinal metastases.

METHODS

Vertebral levels C2 through L5 containing metastases treated with SBRT were included if they were naive to prior surgery or radiation therapy, target delineation was based on consensus guidelines, and 1-year follow-up data were available. Clinical features, including characteristics of the patient, disease, and treatment, were obtained from chart review. Radiomic features were extracted from the planning target volume (PTV) on pretreatment CT and T1-weighted MRI. Clinical and radiomic features selected by least absolute shrinkage and selection operator (LASSO) regression were included in random forest classification models, which were trained to predict VCF within 1 year after SBRT. Model performance was assessed with leave-one-out cross-validation.

RESULTS

Within 1 year after SBRT, 15 of 95 vertebral levels included in the analysis demonstrated new or progressive VCF. Selected clinical features included BMI, performance status, total prescription dose, dose to 99% of the PTV, lumbar location, and 2 components of the Spine Instability Neoplastic Score (SINS): lytic tumor character and spinal misalignment. Selected radiomic features included 5 features from CT and 3 features from MRI. The best-performing classification model, derived from a combination of selected clinical and radiomic features, demonstrated a sensitivity of 0.844, specificity of 0.800, and area under the receiver operating characteristic (ROC) curve (AUC) of 0.878. This model was significantly more accurate than alternative models derived from only selected clinical features (AUC = 0.795, p = 0.048) or only components of the SINS (AUC = 0.579, p < 0.0001).

CONCLUSIONS

In the treatment of spinal metastases with SBRT, a machine learning model incorporating both clinical features and radiomic features from pretreatment imaging predicted VCF at 1 year after SBRT with excellent sensitivity and specificity, outperforming models developed from clinical features or components of the SINS alone. If validated, these findings may allow more judicious selection of patients for prophylactic interventions.

Development and Validation of a Radiomics Model for Differentiating Bone Islands and Osteoblastic Bone Metastases at Abdominal CT

Background It is important to diagnose sclerotic bone lesions in order to determine treatment strategy. Purpose To evaluate the diagnostic performance of a CT radiomics-based machine learning model for differentiating bone islands and osteoblastic bone metastases. Materials and Methods In this retrospective study, patients who underwent contrast-enhanced abdominal CT and were diagnosed with a bone island or osteoblastic metastasis between 2015 to 2019 at either of two different institutions were included: institution 1 for the training set and institution 2 for the external test set. Radiomics features were extracted. The random forest (RF) model was built using 10 selected features, and subsequent 10-fold cross-validation was performed. In the test phase, the RF model was tested with an external test set. Three radiologists reviewed the CT images for the test set. The sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC) were calculated for the models and each of the three radiologists. The AUCs of the radiomics model and radiologists were compared. Results A total of 177 patients (89 with a bone island and 88 with metastasis; mean age, 66 years ± 12 [standard deviation]; 111 men) were in the training set, and 64 (23 with a bone island and 41 with metastasis; mean age, 69 years ± 14; 59 men) were in the test set. Radiomics features (n = 1218) were extracted. The average AUC of the RF model from 10-fold cross-validation was 0.89 (sensitivity, 85% [75 of 88 patients]; specificity, 82% [73 of 89 patients]; and accuracy, 84% [148 of 177 patients]). In the test set, the AUC of the trained RF model was 0.96 (sensitivity, 80% [33 of 41 patients]; specificity, 96% [22 of 23 patients]; and accuracy, 86% [55 of 64 patients]). The AUCs for the three readers were 0.95 (95% CI: 0.90, 1.00), 0.96 (95% CI: 0.90, 1.00), and 0.88 (95% CI: 0.80, 0.96). The AUC of radiomics model was higher than that of only reader 3 (0.96 vs 0.88, respectively; P = .03). Conclusion A CT radiomics-based random forest model was proven useful for differentiating bone islands from osteoblastic metastases and showed better diagnostic performance compared with an inexperienced radiologist. © RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Vannier in this issue.

Radiographic Texture Reproducibility: The Impact of Different Materials, their Arrangement, and Focal Spot Size

Background:

Feature reproducibility is a critical issue in quantitative radiomic studies. The aim of this study is to assess how radiographic radiomic textures behave against changes in phantom materials, their arrangements, and focal spot size.

Method:

A phantom with detachable parts was made using wood, sponge, Plexiglas, and rubber. Each material had 1 cm thickness and was imaged for consecutive time. The phantom also was imaged by change in the arrangement of its materials. Imaging was done with two focal spot sizes including 0.6 and 1.2 mm. All images were acquired with a digital radiography machine. Several texture features were extracted from the same size region of interest in all images. To assess reproducibility, coefficient of variation (COV), intraclass correlation coefficient (ICC), and Bland–Altman tests were used.

Results:

Results show that 59%, 50%, and 4.5% of all features are most reproducible (COV ≤5%) against change in focal spot size, material arrangements, and phantom's materials, respectively. Results on Bland–Altman analysis showed that there is just a nonreproducible feature against change in the focal spot size. On the ICC results, we observed that the ICCs for more features are >0.90 and there were few features with ICC lower than 0.90.

Conclusion:

We showed that radiomic textures are vulnerable against changes in materials, arrangement, and different focal spot sizes. These results suggest that a careful analysis of the effects of these parameters is essential before any radiomic clinical application.

A radiomics platform for computing imaging features from µCT images of Thoroughbred racehorse proximal sesamoid bones: Benchmark performance and evaluation

BACKGROUND

Proximal sesamoid bone (PSB) fractures are the most common fatal musculoskeletal injury in North American racehorses. Computed tomography has the potential to detect morphological changes in bone structure but can be challenging to analyse reliably and quantitatively.

OBJECTIVES

To develop a radiomics platform that allows the comparison of features from micro-CTs (µCT) of PSBs in horses that sustained catastrophic fractures with horses that did not. To compare features calculated with a radiomics approach with features calculated from a previously published study that used quantitative µCT in the same specimens.

STUDY DESIGN

Retrospective study of cadaver specimens of µCT images of PSBs using prospectively applied radiomics.

METHODS

Radiomics features were computed on standardised CT datasets to benchmark the software. Features from µCT images of PSBs from eight horses that sustained PSB fracture and eight controls were computed using the contralateral, intact forelimb from horses sustaining PSB fracture (cases, n = 19) and all available forelimbs for controls (n = 30). Two-hundred and fifteen radiomic features were calculated, and similar or comparable features were compared with those reported in a previous study that used the same specimens.

RESULTS

Morphologic features computed with the radiomics approach, such as volume, minor axis dimensions and anisotropy were highly correlated with previously published data. A high number of imperceptible radiomic features, such as entropy, coarseness and histogram features were also found to be significantly different (P < .01). The extent of the differences in image features for the cases and controls PSBs depends on radiomic calculation settings.

MAIN LIMITATIONS

Only datasets obtained from cadaver specimens were included in the study.

CONCLUSIONS

A radiomics approach for analysing µCT images of PSBs was able to identify and reproduce differences in image features in cases and controls. Furthermore, radiomics revealed many more imperceptible image features between cases and control PSBs.

Radiographic Image Radiomics Feature Reproducibility: A Preliminary Study on the Impact of Field Size

RATIONALE AND OBJECTIVES

Radiomics is an approach to quantifying diseases. Recently, several studies have indicated that radiomics features are vulnerable against imaging parameters. The aim of this study is to assess how radiomics features change with radiographic field sizes, positions in the field size, and mAs.

MATERIALS AND METHODS

A large and small wood phantom and a cotton phantom were prepared and imaged in different field sizes, mAs, and placement in the radiographic field size. A region of interest was drawn on the image features, and twenty two features were extracted. Radiomics feature reproducibility was obtained based on coefficient of variation, Bland-Altman analysis, and intraclass correlation coefficient. Features with coefficient of variation ≤ 5%, intraclass correlation coefficient ≤ 90%, and 1% ≤ U/LRL ≤30% were introduced as robust features. U/LRL is upper/lower reproducibility limits in Bland-Altman.

RESULTS

For all field sizes and all phantoms, features including Difference Variance, Inverse Different Moment, Fraction, Long Run Emphasis, Run Length Non Uniformity, and Short Run Emphasis were found as highly reproducible features. For change in the position of field size, Fraction was the most reproducible in all field sizes and all phantoms. On the mAs change, we found that feature, Short Run Emphasis field 15 × 15 for small wood phantom, and Correlation in all field sizes for Cotton are the most reproducible features.

CONCLUSION

We demonstrated that radiomics features are strongly vulnerable against radiographic field size, positions in the radiation field, mAs, and phantom materials, and reproducibility analyses should be performed before each radiomics study. Moreover, these changing parameters should be considered, and their effects should be minimized in future radiomics studies.

Radiomics for classification of bone mineral loss: A machine learning study

PURPOSE

The purpose of this study was to develop predictive models to classify osteoporosis, osteopenia and normal patients using radiomics and machine learning approaches.

MATERIALS AND METHODS

A total of 147 patients were included in this retrospective single-center study. There were 12 men and 135 women with a mean age of 56.88±10.6 (SD) years (range: 28-87 years). For each patient, seven regions including four lumbar and three femoral including trochanteric, intertrochanteric and neck were segmented on bone mineral densitometry images and 54 texture features were extracted from the regions. The performance of four feature selection methods, including classifier attribute evaluation (CLAE), one rule attribute evaluation (ORAE), gain ratio attribute evaluation (GRAE) and principal components analysis (PRCA) along with four classification methods, including random forest (RF), random committee (RC), K-nearest neighbor (KN) and logit-boost (LB) were evaluated. Four classification categories, including osteopenia vs. normal, osteoporosis vs. normal, osteopenia vs. osteoporosis and osteoporosis+osteopenia vs. osteoporosis were examined for the defined seven regions. The classification model performances were evaluated using the area under the receiver operator characteristic curve (AUC).

RESULTS

The AUC values ranged from 0.50 to 0.78. The combination of methods RF+CLAE, RF+ORAE and RC+ORAE yielded highest performance (AUC=0.78) in discriminating between osteoporosis and normal state in the trochanteric region. The combinations of RF+PRCA and LB+PRCA had the highest performance (AUC=0.76) in discriminating between osteoporosis and normal state in the neck region.

CONCLUSION

The machine learning radiomic approach can be considered as a new method for bone mineral deficiency disease classification using bone mineral densitometry image features.

Radiotherapy dose painting by circadian rhythm based radiomics

Radiotherapy dose painting is a new dose delivery technique to achieve higher treatment outcome. In this approach, does is escalated to high progressive regions which are heterogeneous and determined by advanced medical imaging. Radiomics is issued as a feasible image quantification method to reveal tumor heterogeneity by extraction of high throughput mineable texture features. On the other hand, circadian rhythm is a given biological process that studied as a critical factor to obtain more effective treatment outcome. In this study, we hypothesized that radiotherapy dose painting could be enhanced by using circadian rhythm that is determined on the radiomics maps obtained from medical images. This hypothesis is based on the idea which circadian rhythm could change the tumor heterogeneity and therefore image features.

CT imaging markers to improve radiation toxicity prediction in prostate cancer radiotherapy by stacking regression algorithm

PURPOSE

Radiomic features, clinical and dosimetric factors have the potential to predict radiation-induced toxicity. The aim of this study was to develop prediction models of radiotherapy-induced toxicities in prostate cancer patients based on computed tomography (CT) radiomics, clinical and dosimetric parameters.

METHODS

In this prospective study, prostate cancer patients were included, and radiotherapy-induced urinary and gastrointestinal (GI) toxicities were assessed by Common Terminology Criteria for adverse events. For each patient, clinical and dose volume parameters were obtained. Imaging features were extracted from pre-treatment rectal and bladder wall CT scan of patients. Stacking algorithm and elastic net penalized logistic regression were used in order to feature selection and prediction, simultaneously. The models were fitted by imaging (radiomics model) and clinical/dosimetric (clinical model) features alone and in combinations (clinical-radiomics model). Goodness of fit of the models and performance of classifications were assessed using Hosmer and Lemeshow test, - 2log (likelihood) and area under curve (AUC) of the receiver operator characteristic.

RESULTS

Sixty-four prostate cancer patients were studied, and 33 and 52 patients developed ≥ grade 1 GI and urinary toxicities, respectively. In GI modeling, the AUC for clinical, radiomics and clinical-radiomics models was 0.66, 0.71 and 0.65, respectively. To predict urinary toxicity, the AUC for radiomics, clinical and clinical-radiomics models was 0.71, 0.67 and 0.77, respectively.

CONCLUSIONS

We have shown that CT imaging features could predict radiation toxicities and combination of imaging and clinical/dosimetric features may enhance the predictive performance of radiotoxicity modeling.

Clinical-radiomics nomograms for pre-operative differentiation of sacral chordoma and sacral giant cell tumor based on 3D computed tomography and multiparametric magnetic resonance imaging

OBJECTIVE

To develop and validate clinical-radiomics nomograms based on three-dimensional CT and multiparametric MRI (mpMRI) for pre-operative differentiation of sacral chordoma (SC) and sacral giant cell tumor (SGCT).

METHODS

A total of 83 SC and 54 SGCT patients diagnosed through surgical pathology were retrospectively analyzed. We built six models based on CT, CT enhancement (CTE), T₁ weighted, T₂ weighted, diffusion-weighted imaging (DWI), and contrast-enhanced T₁ weighted features, two radiomics nomograms and two clinical-radiomics nomograms combined radiomics mixed features with clinical data. The area under the receiver operating characteristic curve (AUC) and accuracy (ACC) analysis were used to assess the performance of the models.

RESULTS

SC and SGCT presented significant differences in terms of age, sex, and tumor location (t_age = 9.00, χ²_sex = 10.86, χ²_location = 26.20; p < 0.01). For individual scan, the radiomics model based on diffusion-weighted imaging features yielded the highest AUC of 0.889 and ACC of 0.885, followed by CT (AUC = 0.857; ACC = 0.846) and CT enhancement (AUC = 0.833; ACC = 0.769). For the combined features, the radiomics model based on mixed CT features exhibited a better AUC of 0.942 and ACC of 0.880, whereas mixed MRI features achieved a lower performance than the individual scan. The clinical-radiomics nomogram based on combined CT features achieved the highest AUC of 0.948 and ACC of 0.920.

CONCLUSIONS

The radiomics model based on CT and multiparametricMRI present a certain predictive value in distinguishing SC and SGCT, which can be used for auxiliary diagnosis before operation. The clinical-radiomics nomograms performed better than radiomics nomograms.

ADVANCES IN KNOWLEDGE

Clinical-radiomics nomograms based on CT and mpMRI features can be used for preoperative differentiation of SC and SGCT.

Reject analysis in digital radiography: A local study on radiographers and students' attitude in Iran

Reject analysis is as a quality indicator and critical tool for dose and image quality optimization in radiology departments. By reducing image rejection rate (RR), radiation dose to patients can be reduced effectively, yielding increased total cost-effectiveness. The aims of this study were to assess the rate of image rejection at 2 direct digital radiography (DR) departments to find the sources of rejection and to observe how radiology students and radiographers deal with image rejection.

Two radiology departments were surveyed during a 3-month period for all imaging procedures. Type of examination, numbers, and reasons for digital image rejection were obtained by systems and questionnaire. A predefined questionnaire, including 13 causes for rejection, was filled by radiographers and students.

Out of the 14 022 acquired images, 1116 were rejected, yielding an overall RR of 8%. Highest RRs were found for examination of cervical spine and lumbosacral. Positioning errors and improper patient preparation were the main reasons for digital image rejection.

The image RR was small, but there is a need for optimizing radiographic practice, and enhancing radiographer’s knowledge may enhance the performance.

Radiomics Signature on Computed Tomography Imaging: Association With Lymph Node Metastasis in Patients With Gastric Cancer

Background: To evaluate whether radiomic feature-based computed tomography (CT) imaging signatures allow prediction of lymph node (LN) metastasis in gastric cancer (GC) and to develop a preoperative nomogram for predicting LN status. Methods: We retrospectively analyzed radiomics features of CT images in 1,689 consecutive patients from three cancer centers. The prediction model was developed in the training cohort and validated in internal and external validation cohorts. Lasso regression model was utilized to select features and build radiomics signature. Multivariable logistic regression analysis was utilized to develop the model. We integrated the radiomics signature, clinical T and N stage, and other independent clinicopathologic variables, and this was presented as a radiomics nomogram. The performance of the nomogram was assessed with calibration, discrimination, and clinical usefulness. Results: The radiomics signature was significantly associated with pathological LN stage in training and validation cohorts. Multivariable logistic analysis found the radiomics signature was an independent predictor of LN metastasis. The nomogram showed good discrimination and calibration. Conclusions: The newly developed radiomic signature was a powerful predictor of LN metastasis and the radiomics nomogram could facilitate the preoperative individualized prediction of LN status.

Machine learning-based radiomic models to predict intensity-modulated radiation therapy response, Gleason score and stage in prostate cancer

OBJECTIVE

To develop different radiomic models based on the magnetic resonance imaging (MRI) radiomic features and machine learning methods to predict early intensity-modulated radiation therapy (IMRT) response, Gleason scores (GS) and prostate cancer (Pca) stages.

METHODS

Thirty-three Pca patients were included. All patients underwent pre- and post-IMRT T2-weighted (T2 W) and apparent diffusing coefficient (ADC) MRI. IMRT response was calculated in terms of changes in the ADC value, and patients were divided as responders and non-responders. A wide range of radiomic features from different feature sets were extracted from all T2 W and ADC images. Univariate radiomic analysis was performed to find highly correlated radiomic features with IMRT response, and a paired t test was used to find significant features between responders and non-responders. To find high predictive radiomic models, tenfold cross-validation as the criterion for feature selection and classification was applied on the pre-, post- and delta IMRT radiomic features, and area under the curve (AUC) of receiver operating characteristics was calculated as model performance value.

RESULTS

Of 33 patients, 15 patients (45%) were found as responders. Univariate analysis showed 20 highly correlated radiomic features with IMRT response (20 ADC and 20 T2). Two and fifteen T2 and ADC radiomic features were found as significant (P-value ≤ 0.05) features between responders and non-responders, respectively. Several cross-combined predictive radiomic models were obtained, and post-T2 radiomic models were found as high predictive models (AUC 0.632) followed by pre-ADC (AUC 0.626) and pre-T2 (AUC 0.61). For GS prediction, T2 W radiomic models were found as more predictive (mean AUC 0.739) rather than ADC models (mean AUC 0.70), while for stage prediction, ADC models had higher prediction performance (mean AUC 0.675).

CONCLUSIONS

Radiomic models developed by MR image features and machine learning approaches are noninvasive and easy methods for personalized prostate cancer diagnosis and therapy.