Multi-parametric MRI-based radiomics signature for discriminating between clinically significant and insignificant prostate cancer: Cross-validation of a machine learning method

Artificial Intelligence Based Algorithms for Prostate Cancer Classification and Detection on Magnetic Resonance Imaging: A Narrative Review

Due to the upfront role of magnetic resonance imaging (MRI) for prostate cancer (PCa) diagnosis, a multitude of artificial intelligence (AI) applications have been suggested to aid in the diagnosis and detection of PCa. In this review, we provide an overview of the current field, including studies between 2018 and February 2021, describing AI algorithms for (1) lesion classification and (2) lesion detection for PCa. Our evaluation of 59 included studies showed that most research has been conducted for the task of PCa lesion classification (66%) followed by PCa lesion detection (34%). Studies showed large heterogeneity in cohort sizes, ranging between 18 to 499 patients (median = 162) combined with different approaches for performance validation. Furthermore, 85% of the studies reported on the stand-alone diagnostic accuracy, whereas 15% demonstrated the impact of AI on diagnostic thinking efficacy, indicating limited proof for the clinical utility of PCa AI applications. In order to introduce AI within the clinical workflow of PCa assessment, robustness and generalizability of AI applications need to be further validated utilizing external validation and clinical workflow experiments.

Diagnostic nomogram based on intralesional and perilesional radiomics features and clinical factors of clinically significant prostate cancer

Previous studies on the value of radiomics for diagnosing clinically significant prostate cancer (csPCa) only utilized intralesional features. However, the role of tumor microenvironment is important in tumor generation and progression. The aim of this study is to build and validate a nomogram based on perilesional and intralesional radiomics features and clinical factors for csPCa. This is a retrospective study, which included 140 patients who underwent prostate magnetic resonance imaging (MRI). This study used 3.0T T2-weighted imaging, apparent diffusion coefficient maps (derived from diffusion-weighted images), and dynamic contrast-enhanced MRI. Region of interest (ROI)s were segmented by two radiologists. Intralesional and combined radiomics signatures were built based on radiomics features extracted from intralesional and the combination of radiomics features extracted from intralesional and perilesional volumes. Serum total prostate-specific antigen level and combined radiomics signature scores were used to construct a diagnostic nomogram. Intraclass correlation efficient analysis was used to test intra- and inter-rater agreement of radiomics features. Min-max scalar was used for normalization. One-way analysis of variance or the Mann-Whitney U-test was used for univariate analysis. Receiver operating characteristic curve analysis, accuracy, balanced accuracy, and F1-score were used to evaluate radiomics signatures and the nomogram. Also, the nomogram was evaluated using decision curve analysis in testing cohort. Delong test was used to compare area under the curves (AUCs). Statistical significance was set at p < 0.05. In testing cohort, AUC, accuracy, balanced accuracy, and F1-score of combined radiomics signature (0.94, 0.83, 0.80, and 0.87, respectively) were all higher than that of intralesional radiomics signature (0.90, 0.77, 0.74, and 0.83, respectively). The difference between AUCs was insignificant (p of 0.19). AUC, accuracy, balanced accuracy, and F1-score of the nomogram were 0.96, 0.94, 0.95, and 0.95, respectively. Nomogram was clinically useful when threshold probability of a patient is higher than 0.06. Perilesional radiomics features improved the discrimination ability of the radiomics signature. Diagnostic nomogram had a good performance. LEVEL OF EVIDENCE: 3. TECHNICAL EFFICACY STAGE: 2.

The Primacy of High B-Value 3T-DWI Radiomics in the Prediction of Clinically Significant Prostate Cancer

Predicting clinically significant prostate cancer (csPCa) is crucial in PCa management. 3T-magnetic resonance (MR) systems may have a novel role in quantitative imaging and early csPCa prediction, accordingly. In this study, we develop a radiomic model for predicting csPCa based solely on native b2000 diffusion weighted imaging (DWI_b2000) and debate the effectiveness of apparent diffusion coefficient (ADC) in the same task. In total, 105 patients were retrospectively enrolled between January–November 2020, with confirmed csPCa or ncsPCa based on biopsy. DWI_b2000 and ADC images acquired with a 3T-MRI were analyzed by computing 84 local first-order radiomic features (RFs). Two predictive models were built based on DWI_b2000 and ADC, separately. Relevant RFs were selected through LASSO, a support vector machine (SVM) classifier was trained using repeated 3-fold cross validation (CV) and validated on a holdout set. The SVM models rely on a single couple of uncorrelated RFs (ρ < 0.15) selected through Wilcoxon rank-sum test (p ≤ 0.05) with Holm–Bonferroni correction. On the holdout set, while the ADC model yielded AUC = 0.76 (95% CI, 0.63–0.96), the DWI_b2000 model reached AUC = 0.84 (95% CI, 0.63–0.90), with specificity = 75%, sensitivity = 90%, and informedness = 0.65. This study establishes the primary role of 3T-DWI_b2000 in PCa quantitative analyses, whilst ADC can remain the leading sequence for detection.

Differentiation of small (≤ 3 cm) hepatocellular carcinomas from benign nodules in cirrhotic liver: the added additive value of MRI-based radiomics analysis to LI-RADS version 2018 algorithm

Background

Accurate characterization of small nodules in a cirrhotic liver is challenging. We aimed to determine the additive value of MRI-based radiomics analysis to Liver Imaging Reporting and Data System version 2018 (LI-RADS v 2018) algorithm in differentiating small (≤ 3 cm) hepatocellular carcinomas (HCCs) from benign nodules in cirrhotic liver.

Methods

In this retrospective study, 150 cirrhosis patients with histopathologically confirmed small liver nodules (HCC, 112; benign nodules, 44) were evaluated from January 2013 to October 2018. Based on the LI-RADS algorithm, a LI-RADS category was assigned for each lesion. A radiomics signature was generated based on texture features extracted from T1-weighted, T2W, and apparent diffusion coefficient (ADC) images by using the least absolute shrinkage and selection operator regression model. A nomogram model was developed for the combined diagnosis. Diagnostic performance was assessed using receiver operating characteristic curve (ROC) analysis.

Results

A radiomics signature consisting of eight features was significantly associated with the differentiation of HCCs from benign nodules. Both LI-RADS algorithm (area under ROC [A_z] = 0.898) and the MRI-Based radiomics signature (A_z = 0.917) demonstrated good discrimination, and the nomogram model showed a superior classification performance (A_z = 0.975). Compared with LI-RADS alone, the combined approach significantly improved the specificity (97.7% vs 81.8%, p = 0.030) and positive predictive value (99.1% vs 92.9%, p = 0.031) and afforded comparable sensitivity (97.3% vs 93.8%, p = 0.215) and negative predictive value (93.5% vs 83.7%, p = 0.188).

Conclusions

MRI-based radiomics analysis showed additive value to the LI-RADS v 2018 algorithm for differentiating small HCCs from benign nodules in the cirrhotic liver.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12876-021-01710-y.

Integrated Learning: Screening Optimal Biomarkers for Identifying Preeclampsia in Placental mRNA Samples

Preeclampsia (PE) is a maternal disease that causes maternal and child death. Treatment and preventive measures are not sound enough. The problem of PE screening has attracted much attention. The purpose of this study is to screen placental mRNA to obtain the best PE biomarkers for identifying patients with PE. We use Limma in the R language to screen out the 48 differentially expressed genes with the largest differences and used correlation-based feature selection algorithms to reduce the dimensionality and avoid attribute redundancy arising from too many mRNA samples participating in the classification. After reducing the mRNA attributes, the mRNA samples are sorted from large to small according to information gain. In this study, a classifier model is designed to identify whether samples had PE through mRNA in the placenta. To improve the accuracy of classification and avoid overfitting, three classifiers, including C4.5, AdaBoost, and multilayer perceptron, are used. We use the majority voting strategy integrated with the differentially expressed genes and the genes filtered by the best subset method as comparison methods to train the classifier. The results show that the classification accuracy rate has increased from 79% to 82.2%, and the number of mRNA features has decreased from 48 to 13. This study provides clues for the main PE biomarkers of mRNA in the placenta and provides ideas for the treatment and screening of PE.

A Multi-Center, Multi-Vendor Study to Evaluate the Generalizability of a Radiomics Model for Classifying Prostate cancer: High Grade vs. Low Grade

Radiomics applied in MRI has shown promising results in classifying prostate cancer lesions. However, many papers describe single-center studies without external validation. The issues of using radiomics models on unseen data have not yet been sufficiently addressed. The aim of this study is to evaluate the generalizability of radiomics models for prostate cancer classification and to compare the performance of these models to the performance of radiologists. Multiparametric MRI, photographs and histology of radical prostatectomy specimens, and pathology reports of 107 patients were obtained from three healthcare centers in the Netherlands. By spatially correlating the MRI with histology, 204 lesions were identified. For each lesion, radiomics features were extracted from the MRI data. Radiomics models for discriminating high-grade (Gleason score ≥ 7) versus low-grade lesions were automatically generated using open-source machine learning software. The performance was tested both in a single-center setting through cross-validation and in a multi-center setting using the two unseen datasets as external validation. For comparison with clinical practice, a multi-center classifier was tested and compared with the Prostate Imaging Reporting and Data System version 2 (PIRADS v2) scoring performed by two expert radiologists. The three single-center models obtained a mean AUC of 0.75, which decreased to 0.54 when the model was applied to the external data, the radiologists obtained a mean AUC of 0.46. In the multi-center setting, the radiomics model obtained a mean AUC of 0.75 while the radiologists obtained a mean AUC of 0.47 on the same subset. While radiomics models have a decent performance when tested on data from the same center(s), they may show a significant drop in performance when applied to external data. On a multi-center dataset our radiomics model outperformed the radiologists, and thus, may represent a more accurate alternative for malignancy prediction.

Radiomics and Prostate MRI: Current Role and Future Applications

Multiparametric prostate magnetic resonance imaging (mpMRI) is widely used as a triage test for men at a risk of prostate cancer. However, the traditional role of mpMRI was confined to prostate cancer staging. Radiomics is the quantitative extraction and analysis of minable data from medical images; it is emerging as a promising tool to detect and categorize prostate lesions. In this paper we review the role of radiomics applied to prostate mpMRI in detection and localization of prostate cancer, prediction of Gleason score and PI-RADS classification, prediction of extracapsular extension and of biochemical recurrence. We also provide a future perspective of artificial intelligence (machine learning and deep learning) applied to the field of prostate cancer.

Prediction of Pathological Upgrading at Radical Prostatectomy in Prostate Cancer Eligible for Active Surveillance: A Texture Features and Machine Learning-Based Analysis of Apparent Diffusion Coefficient Maps

Objective

To evaluate a combination of texture features and machine learning-based analysis of apparent diffusion coefficient (ADC) maps for the prediction of Grade Group (GG) upgrading in Gleason score (GS) ≤6 prostate cancer (PCa) (GG1) and GS 3 + 4 PCa (GG2).

Materials and methods

Fifty-nine patients who were biopsy-proven to have GG1 or GG2 and underwent MRI examination with the same MRI scanner prior to transrectal ultrasound (TRUS)-guided systemic biopsy were included. All these patients received radical prostatectomy to confirm the final GG. Patients were divided into training cohort and test cohort. 94 texture features were extracted from ADC maps for each patient. The independent sample t-test or Mann−Whitney U test was used to identify the texture features with statistically significant differences between GG upgrading group and GG non-upgrading group. Texture features of GG1 and GG2 were compared based on the final pathology of radical prostatectomy. We used the least absolute shrinkage and selection operator (LASSO) algorithm to filter features. Four supervised machine learning methods were employed. The prediction performance of each model was evaluated by area under the receiver operating characteristic curve (AUC). The statistical comparison between AUCs was performed.

Results

Six texture features were selected for the machine learning models building. These texture features were significantly different between GG upgrading group and GG non-upgrading group (P < 0.05). The six features had no significant difference between GG1 and GG2 based on the final pathology of radical prostatectomy. All machine learning methods had satisfactory predictive efficacy. The diagnostic performance of nearest neighbor algorithm (NNA) and support vector machine (SVM) was better than random forests (RF) in the training cohort. The AUC, sensitivity, and specificity of NNA were 0.872 (95% CI: 0.750−0.994), 0.967, and 0.778, respectively. The AUC, sensitivity, and specificity of SVM were 0.861 (95%CI: 0.732−0.991), 1.000, and 0.722, respectively. There had no significant difference between AUCs in the test cohort.

Conclusion

A combination of texture features and machine learning-based analysis of ADC maps could predict PCa GG upgrading from biopsy to radical prostatectomy non-invasively with satisfactory predictive efficacy.

Magnetic Resonance Imaging Based Radiomic Models of Prostate Cancer: A Narrative Review

Simple Summary

The increasing interest in implementing artificial intelligence in radiomic models has occurred alongside advancement in the tools used for computer-aided diagnosis. Such tools typically apply both statistical and machine learning methodologies to assess the various modalities used in medical image analysis. Specific to prostate cancer, the radiomics pipeline has multiple facets that are amenable to improvement. This review discusses the steps of a magnetic resonance imaging based radiomics pipeline. Present successes, existing opportunities for refinement, and the most pertinent pending steps leading to clinical validation are highlighted.

Abstract

The management of prostate cancer (PCa) is dependent on biomarkers of biological aggression. This includes an invasive biopsy to facilitate a histopathological assessment of the tumor’s grade. This review explores the technical processes of applying magnetic resonance imaging based radiomic models to the evaluation of PCa. By exploring how a deep radiomics approach further optimizes the prediction of a PCa’s grade group, it will be clear how this integration of artificial intelligence mitigates existing major technological challenges faced by a traditional radiomic model: image acquisition, small data sets, image processing, labeling/segmentation, informative features, predicting molecular features and incorporating predictive models. Other potential impacts of artificial intelligence on the personalized treatment of PCa will also be discussed. The role of deep radiomics analysis-a deep texture analysis, which extracts features from convolutional neural networks layers, will be highlighted. Existing clinical work and upcoming clinical trials will be reviewed, directing investigators to pertinent future directions in the field. For future progress to result in clinical translation, the field will likely require multi-institutional collaboration in producing prospectively populated and expertly labeled imaging libraries.

Associations between Statin/Omega3 Usage and MRI-Based Radiomics Signatures in Prostate Cancer

Prostate cancer is the most common noncutaneous cancer and the second leading cause of cancer deaths among American men. Statins and omega-3 are two medications recently found to correlate with prostate cancer risk and aggressiveness, but the observed associations are complex and controversial. We therefore explore the novel application of radiomics in studying statin and omega-3 usage in prostate cancer patients. On MRIs of 91 prostate cancer patients, two regions of interest (ROIs), the whole prostate and the peripheral region of the prostate, were manually segmented. From each ROI, 944 radiomic features were extracted after field bias correction and normalization. Heatmaps were generated to study the radiomic feature patterns against statin or omega-3 usage. Radiomics models were trained on selected features and evaluated with 500-round threefold cross-validation for each drug/ROI combination. On the 1500 validation datasets, the radiomics model achieved average AUCs of 0.70, 0.74, 0.78, and 0.72 for omega-3/prostate, omega-3/peripheral, statin/prostate, and statin/peripheral, respectively. As the first study to analyze radiomics in relation to statin and omega-3 uses in prostate cancer patients, our study preliminarily established the existence of imaging-identifiable tissue-level changes in the prostate and illustrated the potential usefulness of radiomics for further exploring these medications’ effects and mechanisms in prostate cancer.

LMTK2 as Potential Biomarker for Stratification between Clinically Insignificant and Clinically Significant Prostate Cancer

A set of prostate tumors tend to grow slowly and do not require active treatment. Therefore, stratification between patients with clinically significant and clinically insignificant prostate cancer (PC) remains a vital issue to avoid overtreatment. Fast development of genetic technologies accelerated development of next-generation molecular tools for reliable PC diagnosis. The aim of this study is to evaluate the diagnostic value of molecular biomarkers (CRISP3, LMTK2, and MSMB) for separation of PC cases from benign prostatic changes and more specifically for identification of clinically significant PC from all pool of PC cases in patients with rising PSA levels. Patients (n = 200) who had rising PSA (PSA II) after negative transrectal systematic prostate biopsy due to elevated PSA (PSA I) were eligible to the study. In addition to PSA concentration, PSA density was calculated for each patient. Gene expression level was measured in peripheral blood samples of cases applying RT-PCR, while MSMB (−57 C/T) polymorphism was identified by pyrosequencing. LMTK2 and MSMB significantly differentiated control group from both BPD and PC groups. MSMB expression tended to increase from the major alleles of the CC genotype to the minor alleles of the TT genotype. PSA density was the only clinical characteristic that significantly differentiated clinically significant PC from clinically insignificant PC. Therefore, LMTK2 expression and PSA density were significantly distinguished between clinically significant PC and clinically insignificant PC. PSA density rather than PSA can differentiate PC from the benign prostate disease and, in combination with LMTK2, assist in stratification between clinically insignificant and clinically significant PC.

A radiomics machine learning-based redefining score robustly identifies clinically significant prostate cancer in equivocal PI-RADS score 3 lesions

PURPOSE

PI-RADS score 3 is recognized as equivocal likelihood of clinically significant prostate cancer (csPCa) occurrence. We aimed to develop a Radiomics machine learning (RML)-based redefining score to screen out csPCa in equivocal PI-RADS score 3 category.

METHODS

Total of 263 patients with the dominant index lesion scored PI-RADS 3 who underwent biopsy and/or follow-up formed the primary cohort. One-step RML (RML-i) model integrated radiomic features of T2WI, DWI, and ADC images all together, and two-step RML (RML-ii) model integrated the three independent radiomic signatures from T2WI (T2WI_RS), DWI (DWI_RS), and ADC (ADC_RS) separately into a regression model. The two RML models, as well as T2WI_RS, DWI_RS, and ADC_RS, were compared using the receiver operating characteristic-derived area under the curve (AUC), calibration plot, and decision-curve analysis (DCA). Two radiologists were asked to give a subjective binary assessment, and Cohen's kappa statistics were calculated.

RESULTS

A total of 59/263 (22.4%) csPCa were identified. Inter-reader agreement was moderate (Kappa = 0.435). The AUC of RML-i (0.89; 95% CI 0.88-0.90) is higher (p = 0.003) than that of RML-ii (0.87; 95% CI 0.86-0.88). The DCA demonstrated that the RML-i and RML-ii significantly improved risk prediction at threshold probabilities of csPCa at 20% to 80% compared with doing-none or doing-all by PI-RADS score 3 or stratifying by separated DWI_RS, ADC_RS, or T2WI_RS.

CONCLUSION

Our RML models have the potential to predict csPCa in PI-RADS score 3 lesions, thus can inform the decision making process of biopsy.

The role of radiomics in prostate cancer radiotherapy

"Radiomics," as it refers to the extraction and analysis of a large number of advanced quantitative radiological features from medical images using high-throughput methods, is perfectly suited as an engine for effectively sifting through the multiple series of prostate images from before, during, and after radiotherapy (RT). Multiparametric (mp)MRI, planning CT, and cone beam CT (CBCT) routinely acquired throughout RT and the radiomics pipeline are developed for extraction of thousands of variables. Radiomics data are in a format that is appropriate for building descriptive and predictive models relating image features to diagnostic, prognostic, or predictive information. Prediction of Gleason score, the histopathologic cancer grade, has been the mainstay of the radiomic efforts in prostate cancer. While Gleason score (GS) is still the best predictor of treatment outcome, there are other novel applications of quantitative imaging that are tailored to RT. In this review, we summarize the radiomics efforts and discuss several promising concepts such as delta-radiomics and radiogenomics for utilizing image features for assessment of the aggressiveness of prostate cancer and its outcome. We also discuss opportunities for quantitative imaging with the advance of instrumentation in MRI-guided therapies.

A Quality Control System for Automated Prostate Segmentation on T2-Weighted MRI

Computer-aided detection and diagnosis (CAD) systems have the potential to improve robustness and efficiency compared to traditional radiological reading of magnetic resonance imaging (MRI). Fully automated segmentation of the prostate is a crucial step of CAD for prostate cancer, but visual inspection is still required to detect poorly segmented cases. The aim of this work was therefore to establish a fully automated quality control (QC) system for prostate segmentation based on T2-weighted MRI. Four different deep learning-based segmentation methods were used to segment the prostate for 585 patients. First order, shape and textural radiomics features were extracted from the segmented prostate masks. A reference quality score (QS) was calculated for each automated segmentation in comparison to a manual segmentation. A least absolute shrinkage and selection operator (LASSO) was trained and optimized on a randomly assigned training dataset (N = 1756, 439 cases from each segmentation method) to build a generalizable linear regression model based on the radiomics features that best estimated the reference QS. Subsequently, the model was used to estimate the QSs for an independent testing dataset (N = 584, 146 cases from each segmentation method). The mean ± standard deviation absolute error between the estimated and reference QSs was 5.47 ± 6.33 on a scale from 0 to 100. In addition, we found a strong correlation between the estimated and reference QSs (rho = 0.70). In conclusion, we developed an automated QC system that may be helpful for evaluating the quality of automated prostate segmentations.

Advanced zoomed diffusion-weighted imaging vs. full-field-of-view diffusion-weighted imaging in prostate cancer detection: a radiomic features study

OBJECTIVES

We aimed to compare the efficiency of prostate cancer (PCa) detection using a radiomics signature based on advanced zoomed diffusion-weighted imaging and conventional full-field-of-view DWI.

METHODS

A total of 136 patients, including 73 patients with PCa and 63 without PCa, underwent multi-parametric magnetic resonance imaging (mp-MRI). Radiomic features were extracted from prostate lesion areas segmented on full-field-of-view DWI with b-value = 1500 s/mm² (f-DWI_b1500), advanced zoomed DWI images with b-value = 1500 s/mm² (z-DWI_b1500), calculated zoomed DWI with b-value = 2000 s/mm² (z-calDWI_b2000), and apparent diffusion coefficient (ADC) maps derived from both sequences (f-ADC and z-ADC). Single-imaging modality radiomics signature, mp-MRI radiomics signature, and a mixed model based on mp-MRI and clinically independent risk factors were built to predict PCa probability. The diagnostic efficacy and the potential net benefits of each model were evaluated.

RESULTS

Both z-DWI_b1500 and z-calDWI_b2000 had significantly better predictive performance than f-DWIb1500 (z-DWIb1500 vs. f-DWIb1500: p = 0.048; z-calDWIb2000 vs. f-DWIb1500: p = 0.014). z-ADC had a slightly higher area under the curve (AUC) value compared with f-ADC value but was not significantly different (p = 0.127). For predicting the presence of PCa, the AUCs of clinical independent risk factors model, mp-MRI model, and mixed model were 0.81, 0.93, and 0.94 in training sets, and 0.74, 0.92, and 0.93 in validation sets, respectively.

CONCLUSION

Radiomics signatures based on the z-DWI technology had better diagnostic accuracy for PCa than that based on the f-DWI technology. The mixed model was better at diagnosing PCa and guiding clinical interventions for patients with suspected PCa compared with mp-MRI signatures and clinically independent risk factors.

KEY POINTS

• Advanced zoomed DWI technology can improve the diagnostic accuracy of radiomics signatures for PCa. • Radiomics signatures based on z-calDWI_b2000 have the best diagnostic performance among individual imaging modalities. • Compared with the independent clinical risk factors and the mp-MRI model, the mixed model has the best diagnostic efficiency.

The Application of Artificial Intelligence in Prostate Cancer Management—What Improvements Can Be Expected? A Systematic Review

Artificial Intelligence (AI) is progressively remodeling our daily life. A large amount of information from “big data” now enables machines to perform predictions and improve our healthcare system. AI has the potential to reshape prostate cancer (PCa) management thanks to growing applications in the field. The purpose of this review is to provide a global overview of AI in PCa for urologists, pathologists, radiotherapists, and oncologists to consider future changes in their daily practice. A systematic review was performed, based on PubMed MEDLINE, Google Scholar, and DBLP databases for original studies published in English from January 2009 to January 2019 relevant to PCa, AI, Machine Learning, Artificial Neural Networks, Convolutional Neural Networks, and Natural-Language Processing. Only articles with full text accessible were considered. A total of 1008 articles were reviewed, and 48 articles were included. AI has potential applications in all fields of PCa management: analysis of genetic predispositions, diagnosis in imaging, and pathology to detect PCa or to differentiate between significant and non-significant PCa. AI also applies to PCa treatment, whether surgical intervention or radiotherapy, skills training, or assessment, to improve treatment modalities and outcome prediction. AI in PCa management has the potential to provide a useful role by predicting PCa more accurately, using a multiomic approach and risk-stratifying patients to provide personalized medicine.

FeAture Explorer (FAE): A tool for developing and comparing radiomics models

In radiomics studies, researchers usually need to develop a supervised machine learning model to map image features onto the clinical conclusion. A classical machine learning pipeline consists of several steps, including normalization, feature selection, and classification. It is often tedious to find an optimal pipeline with appropriate combinations. We designed an open-source software package named FeAture Explorer (FAE). It was programmed with Python and used NumPy, pandas, and scikit-learning modules. FAE can be used to extract image features, preprocess the feature matrix, develop different models automatically, and evaluate them with common clinical statistics. FAE features a user-friendly graphical user interface that can be used by radiologists and researchers to build many different pipelines, and to compare their results visually. To prove the effectiveness of FAE, we developed a candidate model to classify the clinical-significant prostate cancer (CS PCa) and non-CS PCa using the PROSTATEx dataset. We used FAE to try out different combinations of feature selectors and classifiers, compare the area under the receiver operating characteristic curve of different models on the validation dataset, and evaluate the model using independent test data. The final model with the analysis of variance as the feature selector and linear discriminate analysis as the classifier was selected and evaluated conveniently by FAE. The area under the receiver operating characteristic curve on the training, validation, and test dataset achieved results of 0.838, 0.814, and 0.824, respectively. FAE allows researchers to build radiomics models and evaluate them using an independent testing dataset. It also provides easy model comparison and result visualization. We believe FAE can be a convenient tool for radiomics studies and other medical studies involving supervised machine learning.

Test-retest repeatability of a deep learning architecture in detecting and segmenting clinically significant prostate cancer on apparent diffusion coefficient (ADC) maps

OBJECTIVES

To evaluate short-term test-retest repeatability of a deep learning architecture (U-Net) in slice- and lesion-level detection and segmentation of clinically significant prostate cancer (csPCa: Gleason grade group > 1) using diffusion-weighted imaging fitted with monoexponential function, ADC_m.

METHODS

One hundred twelve patients with prostate cancer (PCa) underwent 2 prostate MRI examinations on the same day. PCa areas were annotated using whole mount prostatectomy sections. Two U-Net-based convolutional neural networks were trained on three different ADC_m b value settings for (a) slice- and (b) lesion-level detection and (c) segmentation of csPCa. Short-term test-retest repeatability was estimated using intra-class correlation coefficient (ICC(3,1)), proportionate agreement, and dice similarity coefficient (DSC). A 3-fold cross-validation was performed on training set (N = 78 patients) and evaluated for performance and repeatability on testing data (N = 34 patients).

RESULTS

For the three ADC_m b value settings, repeatability of mean ADC_m of csPCa lesions was ICC(3,1) = 0.86-0.98. Two CNNs with U-Net-based architecture demonstrated ICC(3,1) in the range of 0.80-0.83, agreement of 66-72%, and DSC of 0.68-0.72 for slice- and lesion-level detection and segmentation of csPCa. Bland-Altman plots suggest that there is no systematic bias in agreement between inter-scan ground truth segmentation repeatability and segmentation repeatability of the networks.

CONCLUSIONS

For the three ADC_m b value settings, two CNNs with U-Net-based architecture were repeatable for the problem of detection of csPCa at the slice-level. The network repeatability in segmenting csPCa lesions is affected by inter-scan variability and ground truth segmentation repeatability and may thus improve with better inter-scan reproducibility.

KEY POINTS

• For the three ADC_m b value settings, two CNNs with U-Net-based architecture were repeatable for the problem of detection of csPCa at the slice-level. • The network repeatability in segmenting csPCa lesions is affected by inter-scan variability and ground truth segmentation repeatability and may thus improve with better inter-scan reproducibility.

Development of a Novel, Multi-Parametric, MRI-Based Radiomic Nomogram for Differentiating Between Clinically Significant and Insignificant Prostate Cancer

Objectives: To develop and validate a predictive model for discriminating clinically significant prostate cancer (csPCa) from clinically insignificant prostate cancer (ciPCa).

Methods: This retrospective study was performed with 159 consecutively enrolled pathologically confirmed PCa patients from two medical centers. The dataset was allocated to a training group (n = 54) and an internal validation group (n = 22) from one center along with an external independent validation group (n = 83) from another center. A total of 1,188 radiomic features were extracted from T2WI, diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) images derived from DWI for each patient. Multivariable logistic regression analysis was performed to develop the model, incorporating the radiomic signature, ADC value, and independent clinical risk factors. This was presented using a radiomic nomogram. The receiver operating characteristic (ROC) curve was utilized to assess the predictive efficacy of the radiomic nomogram in both the training and validation groups. The decision curve analysis was used to evaluate which model achieved the most net benefit.

Results: The radiomic signature, which was made up of 10 selected features, was significantly associated with csPCa (P < 0.001 for both training and internal validation groups). The area under the curve (AUC) values of discriminating csPCa for the radiomics signature were 0.95 (training group), 0.86 (internal validation group), and 0.81 (external validation group). Multivariate logistic analysis identified the radiomic signature and ADC value as independent parameters of predicting csPCa. Then, the combination nomogram incorporating the radiomic signature and ADC value demonstrated a favorable classification capability with the AUC of 0.95 (training group), 0.93 (internal validation group), and 0.84 (external validation group). Appreciable clinical utility of this model was illustrated using the decision curve analysis for the nomogram.

Conclusions: The nomogram, incorporating radiomic signature and ADC value, provided an individualized, potential approach for discriminating csPCa from ciPCa.

Automated Classification of Significant Prostate Cancer on MRI: A Systematic Review on the Performance of Machine Learning Applications

Significant prostate carcinoma (sPCa) classification based on MRI using radiomics or deep learning approaches has gained much interest, due to the potential application in assisting in clinical decision-making.

OBJECTIVE

To systematically review the literature (i) to determine which algorithms are most frequently used for sPCa classification, (ii) to investigate whether there exists a relation between the performance and the method or the MRI sequences used, (iii) to assess what study design factors affect the performance on sPCa classification, and (iv) to research whether performance had been evaluated in a clinical setting Methods: The databases Embase and Ovid MEDLINE were searched for studies describing machine learning or deep learning classification methods discriminating between significant and nonsignificant PCa on multiparametric MRI that performed a valid validation procedure. Quality was assessed by the modified radiomics quality score. We computed the median area under the receiver operating curve (AUC) from overall methods and the interquartile range.

RESULTS

From 2846 potentially relevant publications, 27 were included. The most frequent algorithms used in the literature for PCa classification are logistic regression (22%) and convolutional neural networks (CNNs) (22%). The median AUC was 0.79 (interquartile range: 0.77-0.87). No significant effect of number of included patients, image sequences, or reference standard on the reported performance was found. Three studies described an external validation and none of the papers described a validation in a prospective clinical trial.

CONCLUSIONS

To unlock the promising potential of machine and deep learning approaches, validation studies and clinical prospective studies should be performed with an established protocol to assess the added value in decision-making.

Prostate MRI radiomics: A systematic review and radiomic quality score assessment

BACKGROUND

Radiomics have the potential to further increase the value of MRI in prostate cancer management. However, implementation in clinical practice is still far and concerns have been raised regarding the methodological quality of radiomic studies. Therefore, we aimed to systematically review the literature to assess the quality of prostate MRI radiomic studies using the radiomics quality score (RQS).

METHODS

Multiple medical literature archives (PubMed, Web of Science and EMBASE) were searched to retrieve original investigations focused on prostate MRI radiomic approaches up to the end of June 2019. Three researchers independently assessed each paper using the RQS. Data from the most experienced researcher were used for descriptive analysis. Inter-rater reproducibility was assessed using the intraclass correlation coefficient (ICC) on the total RQS score.

RESULTS

73 studies were included in the analysis. Overall, the average RQS total score was 7.93 ± 5.13 on a maximum of 36 points, with a final average percentage of 23 ± 13%. Among the most critical items, the lack of feature robustness testing strategies and external validation datasets. The ICC resulted poor to moderate, with an average value of 0.57 and 95% Confidence Intervals between 0.44 and 0.69.

CONCLUSIONS

Current studies on prostate MRI radiomics still lack the quality required to allow their introduction in clinical practice.

Differential Diagnosis of Benign and Malignant Thyroid Nodules Using Deep Learning Radiomics of Thyroid Ultrasound Images

PURPOSE

We aimed to propose a highly automatic and objective model named deep learning Radiomics of thyroid (DLRT) for the differential diagnosis of benign and malignant thyroid nodules from ultrasound (US) images.

METHODS

We retrospectively enrolled and finally include US images and fine-needle aspiration biopsies from 1734 patients with 1750 thyroid nodules. A basic convolutional neural network (CNN) model, a transfer learning (TL) model, and a newly designed model named deep learning Radiomics of thyroid (DLRT) were used for the investigation. Their diagnostic accuracy was further compared with human observers (one senior and one junior US radiologist). Moreover, the robustness of DLRT over different US instruments was also validated. Analysis of receiver operating characteristic (ROC) curves were performed to calculate optimal area under it (AUC) for benign and malignant nodules. One observer helped to delineate the nodules.

RESULTS

AUCs of DLRT were 0.96 (95% confidence interval [CI]: 0.94-0.98), 0.95 (95% confidence interval [CI]: 0.93-0.97) and 0.97 (95% confidence interval [CI]: 0.95-0.99) in the training, internal and external validation cohort, respectively, which were significantly better than other deep learning models (P < 0.01) and human observers (P < 0.001). No significant difference was found when applying DLRT on thyroid US images acquired from different US instruments.

CONCLUSIONS

DLRT shows the best overall performance comparing with other deep learning models and human observers. It holds great promise for improving the differential diagnosis of benign and malignant thyroid nodules.

Radiomics with artificial intelligence: a practical guide for beginners

Radiomics is a relatively new word for the field of radiology, meaning the extraction of a high number of quantitative features from medical images. Artificial intelligence (AI) is broadly a set of advanced computational algorithms that basically learn the patterns in the data provided to make predictions on unseen data sets. Radiomics can be coupled with AI because of its better capability of handling a massive amount of data compared with the traditional statistical methods. Together, the primary purpose of these fields is to extract and analyze as much and meaningful hidden quantitative data as possible to be used in decision support. Nowadays, both radiomics and AI have been getting attention for their remarkable success in various radiological tasks, which has been met with anxiety by most of the radiologists due to the fear of replacement by intelligent machines. Considering ever-developing advances in computational power and availability of large data sets, the marriage of humans and machines in future clinical practice seems inevitable. Therefore, regardless of their feelings, the radiologists should be familiar with these concepts. Our goal in this paper was three-fold: first, to familiarize radiologists with the radiomics and AI; second, to encourage the radiologists to get involved in these ever-developing fields; and, third, to provide a set of recommendations for good practice in design and assessment of future works.

[Artificial intelligence and radiomics in MRI-based prostate diagnostics]

CLINICAL/METHODICAL ISSUE

In view of the diagnostic complexity and the large number of examinations, modern radiology is challenged to identify clinically significant prostate cancer (PCa) with high sensitivity and specificity. Meanwhile overdiagnosis and overtreatment of clinically nonsignificant carcinomas need to be avoided.

STANDARD RADIOLOGICAL METHODS

Increasingly, international guidelines recommend multiparametric magnetic resonance imaging (mpMRI) as first-line investigation in patients with suspected PCa.

METHODICAL INNOVATIONS

Image interpretation according to the PI-RADS criteria is limited by interobserver variability. Thus, rapid developments in the field of automated image analysis tools, including radiomics and artificial intelligence (AI; machine learning, deep learning), give hope for further improvement in patient care.

PERFORMANCE

AI focuses on the automated detection and classification of PCa, but it also attempts to stratify tumor aggressiveness according to the Gleason score. Recent studies present good to very good results in radiomics or AI-supported mpMRI diagnosis. Nevertheless, these systems are not widely used in clinical practice.

ACHIEVEMENTS AND PRACTICAL RECOMMENDATIONS

In order to apply these innovative technologies, a growing awareness for the need of structured data acquisition, development of robust systems and an increased acceptance of AI as diagnostic support are needed. If AI overcomes these obstacles, it may play a key role in the quantitative and reproducible image-based diagnosis of ever-increasing prostate MRI examination volumes.

The performance of intravoxel-incoherent motion diffusion-weighted imaging derived hypoxia for the risk stratification of prostate cancer in peripheral zone

PURPOSE

To assess the association between intravoxel-incoherent motion diffusion-weighted imaging (IVIM) derived hypoxia and the aggressiveness of prostate cancer (PCa) and to explore its contribution to the risk stratification of PCa.

METHODS

Seventy-five peripheral zone (PZ) PCa patients, who underwent multiparametric MRI (mpMRI), were included in this study. Systematic ultrasound guided biopsy was used as reference. IVIM was acquired with 5 b values (b = 0∼750 s/mm²). Apparent diffusion coefficient (ADC), pure tissue diffusion (Ds), volume fraction of pseudo-diffusion (fp), hypoxic fraction (HF_DWI), hypoxia score (HS_DWI) and relative oxygen saturation(rOS_DWI), were calculated and histogram analysis was applied. Groups comparison was performed between low-intermediate-grade group (LG, the International Society of Urological Pathology (ISUP) Gleason Grade (GG) ≤2) and high-grade (HG, ISUP GG ≥ 3) group. The correlation between diffusion parameters and ISUP GG was assessed. Cross-validated Support Vector Machine (SVM) Classification was performed and compared with univariate ROC analysis to explore the risk stratification of PZ PCa.

RESULTS

Mean, median, and the 10th percentile of Ds differed significantly between groups (p < 0.05). Several parameters significantly correlated with ISUP grade, and the 10th percentile of Ds showed the strongest correlation (ρ= - 0.284). The prediction model containing IVIM derived hypoxia yielded an area under the receiver operating characteristics curve (AUC) ranging 0.749-0.786 for cross-validation. The AUCs of the SVM modeling were higher than that of any single parameter.

CONCLUSION

IVIM derived hypoxia demonstrated significant correlation with the aggressiveness of PCa. It's supplemental to the MRI assessment of PCa with a promising stratification of risk stratification of PZ PCa.

Online Transfer Learning for Differential Diagnosis of Benign and Malignant Thyroid Nodules With Ultrasound Images

OBJECTIVE

We aimed to propose a highly automatic and objective model named online transfer learning (OTL) for the differential diagnosis of benign and malignant thyroid nodules from ultrasound (US) images.

METHODS

The OTL mothed combined the strategy of transfer learning and online learning. Two datasets (1750 thyroid nodules with 1078 benign and 672 malignant nodules, and 3852 thyroid nodules with 3213 benign and 639 malignant nodules) were collected to develop the model. The diagnostic accuracy was also compared with VGG-16 based transfer learning model and different input images based model. Analysis of receiver operating characteristic (ROC) curves were performed to calculate optimal area under it (AUC) for benign and malignant nodules.

RESULTS

AUC, sensitivity and specificity of OTL were 0.98 (95% confidence interval [CI]: 0.97-0.99), 98.7% (95% confidence interval [CI]: 97.8%-99.6%) and 98.8% (95% confidence interval [CI]: 97.9%-99.7%) in the final online learning step, which was significantly better than other deep learning models (P < 0.01).

CONCLUSION

OTL model shows the best overall performance comparing with other deep learning models. The model holds a good potential for improving the overall diagnostic efficacy in thyroid nodule US examinations.

SIGNIFICANCE

The proposed OTL model could be seamlessly integrated into the conventional work-flow of thyroid nodule US examinations.

T2-Weighted Image-Based Radiomics Signature for Discriminating Between Seminomas and Nonseminoma

Objective: To evaluate the performance of a T2-weighted image (T2WI)-based radiomics signature for differentiating between seminomas and nonseminomas.

Materials and Methods: In this retrospective study, 39 patients with testicular germ-cell tumors (TGCTs) confirmed by radical orchiectomy were enrolled, including 19 cases of seminomas and 20 cases of nonseminomas. All patients underwent 3T magnetic resonance imaging (MRI) before radical orchiectomy. Eight hundred fifty-one radiomics features were extracted from the T2WI of each patient. Intra- and interclass correlation coefficients were used to select the features with excellent stability and repeatability. Then, we used the minimum-redundancy maximum-relevance (mRMR) and the least absolute shrinkage and selection operator (LASSO) algorithms for feature selection and radiomics signature development. Receiver operating characteristic curve analysis was used to evaluate the diagnostic performance of the radiomics signature.

Results: Five features were selected to build the radiomics signature. The radiomics signature was significantly different between the seminomas and nonseminomas (p < 0.01). The area under the curve (AUC), sensitivity, and specificity of the radiomics signature for discriminating between seminomas and nonseminomas were 0.979 (95% CI: 0.873–1.000), 90.00 (95% CI: 68.3–98.8), and 100.00 (95% CI: 82.4–100.0), respectively.

Conclusion: The T2WI-based radiomics signature has the potential to non-invasively discriminate between seminomas and nonseminomas.