Machine and Deep Learning Prediction Of Prostate Cancer Aggressiveness Using Multiparametric MRI

Prostate cancer (PCa) is the most frequent male malignancy and the assessment of PCa aggressiveness, for which a biopsy is required, is fundamental for patient management. Currently, multiparametric (mp) MRI is strongly recommended before biopsy. Quantitative assessment of mpMRI might provide the radiologist with an objective and noninvasive tool for supporting the decision-making in clinical practice and decreasing intra- and inter-reader variability. In this view, high dimensional radiomics features and Machine Learning (ML) techniques, along with Deep Learning (DL) methods working on raw images directly, could assist the radiologist in the clinical workflow. The aim of this study was to develop and validate ML/DL frameworks on mpMRI data to characterize PCas according to their aggressiveness. We optimized several ML/DL frameworks on T2w, ADC and T2w+ADC data, using a patient-based nested validation scheme. The dataset was composed of 112 patients (132 peripheral lesions with Prostate Imaging Reporting and Data System (PI-RADS) score ≥ 3) acquired following both PI-RADS 2.0 and 2.1 guidelines. Firstly, ML/DL frameworks trained and validated on PI-RADS 2.0 data were tested on both PI-RADS 2.0 and 2.1 data. Then, we trained, validated and tested ML/DL frameworks on a multi PI-RADS dataset. We reported the performances in terms of Area Under the Receiver Operating curve (AUROC), specificity and sensitivity. The ML/DL frameworks trained on T2w data achieved the overall best performance. Notably, ML and DL frameworks trained and validated on PI-RADS 2.0 data obtained median AUROC values equal to 0.750 and 0.875, respectively, on unseen PI-RADS 2.0 test set. Similarly, ML/DL frameworks trained and validated on multi PI-RADS T2w data showed median AUROC values equal to 0.795 and 0.750, respectively, on unseen multi PI-RADS test set. Conversely, all the ML/DL frameworks trained and validated on PI-RADS 2.0 data, achieved AUROC values no better than the chance level when tested on PI-RADS 2.1 data. Both ML/DL techniques applied on mpMRI seem to be a valid aid in predicting PCa aggressiveness. In particular, ML/DL frameworks fed with T2w images data (objective, fast and non-invasive) show good performances and might support decision-making in patient diagnostic and therapeutic management, reducing intra- and inter-reader variability.

Download Full-text

An integrated nomogram combining deep learning, Prostate Imaging–Reporting and Data System (PI-RADS) scoring, and clinical variables for identification of clinically significant prostate cancer on biparametric MRI: a retrospective multicentre study

The Lancet Digital Health ◽

10.1016/s2589-7500(21)00082-0 ◽

2021 ◽

Vol 3 (7) ◽

pp. e445-e454

Author(s):

Amogh Hiremath ◽

Rakesh Shiradkar ◽

Pingfu Fu ◽

Amr Mahran ◽

Ardeshir R Rastinehad ◽

...

Keyword(s):

Prostate Cancer ◽

Deep Learning ◽

Multicentre Study ◽

Data System ◽

Clinical Variables ◽

Prostate Imaging ◽

Clinically Significant

Download Full-text

Biofluid quantification of TWEAK/Fn14 axis in combination with a selected biomarker panel improves assessment of prostate cancer aggressiveness

Journal of Translational Medicine ◽

10.1186/s12967-019-2053-6 ◽

2019 ◽

Vol 17 (1) ◽

Cited By ~ 2

Author(s):

Xavier Ruiz-Plazas ◽

Esther Rodríguez-Gallego ◽

Marta Alves ◽

Antonio Altuna-Coy ◽

Javier Lozano-Bartolomé ◽

...

Keyword(s):

Prostate Cancer ◽

Clinical Biochemistry ◽

Specific Antigen ◽

Total Serum ◽

Mrna Levels ◽

Biomarker Panel ◽

Non Invasive ◽

Clinical Biomarkers ◽

Cancer Aggressiveness ◽

Aggressive Tumors

Abstract Background Conventional clinical biomarkers cannot accurately differentiate indolent from aggressive prostate cancer (PCa). We investigated the usefulness of a biomarker panel measured exclusively in biofluids for assessment of PCa aggressiveness. Methods We collected biofluid samples (plasma/serum/semen/post-prostatic massage urine) from 98 patients that had undergone radical prostatectomy. Clinical biochemistry was performed and several cytokines/chemokines including soluble(s) TWEAK, sFn14, sCD163, sCXCL5 and sCCL7 were quantified by ELISA in selected biofluids. Also, the expression of KLK2, KLK3, Fn14, CD163, CXCR2 and CCR3 was quantified by real-time PCR in semen cell sediment. Univariate, logistic regression, and receiver operating characteristic (ROC) analyses were used to assess the predictive ability of the selected biomarker panel in conjunction with clinical and metabolic variables for the evaluation of PCa aggressiveness. Results Total serum levels of prostate-specific antigen (PSA), semen levels of sTWEAK, fasting glycemia and mRNA levels of Fn14, KLK2, CXCR2 and CCR3 in semen cell sediment constituted a panel of markers that was significantly different between patients with less aggressive tumors [International Society of Urological Pathology (ISUP) grade I and II] and those with more aggressive tumors (ISUP grade III, IV and V). ROC curve analysis showed that this panel could be used to correctly classify tumor aggressiveness in 90.9% of patients. Area under the curve (AUC) analysis revealed that this combination was more accurate [AUC = 0.913 95% confidence interval (CI) 0.782–1] than a classical non-invasive selected clinical panel comprising age, tumor clinical stage (T-classification) and total serum PSA (AUC = 0.721 95% CI 0.613–0.830). Conclusions TWEAK/Fn14 axis in combination with a selected non-invasive biomarker panel, including conventional clinical biochemistry, can improve the predictive power of serum PSA levels and could be used to classify PCa aggressiveness.

Download Full-text

IGRNet: A Deep Learning Model for Non-Invasive, Real-Time Diagnosis of Prediabetes through Electrocardiograms

Sensors ◽

10.3390/s20092556 ◽

2020 ◽

Vol 20 (9) ◽

pp. 2556

Author(s):

Liyang Wang ◽

Yao Mu ◽

Jing Zhao ◽

Xiaoya Wang ◽

Huilian Che

Keyword(s):

Machine Learning ◽

Deep Learning ◽

Real Time ◽

Clinical Symptoms ◽

Characteristic Curve ◽

Learning Model ◽

Machine Learning Algorithms ◽

Test Set ◽

Non Invasive ◽

Deep Learning Model

The clinical symptoms of prediabetes are mild and easy to overlook, but prediabetes may develop into diabetes if early intervention is not performed. In this study, a deep learning model—referred to as IGRNet—is developed to effectively detect and diagnose prediabetes in a non-invasive, real-time manner using a 12-lead electrocardiogram (ECG) lasting 5 s. After searching for an appropriate activation function, we compared two mainstream deep neural networks (AlexNet and GoogLeNet) and three traditional machine learning algorithms to verify the superiority of our method. The diagnostic accuracy of IGRNet is 0.781, and the area under the receiver operating characteristic curve (AUC) is 0.777 after testing on the independent test set including mixed group. Furthermore, the accuracy and AUC are 0.856 and 0.825, respectively, in the normal-weight-range test set. The experimental results indicate that IGRNet diagnoses prediabetes with high accuracy using ECGs, outperforming existing other machine learning methods; this suggests its potential for application in clinical practice as a non-invasive, prediabetes diagnosis technology.

Download Full-text

In1-ghrelin splicing variant as a key element in the pathophysiological association between obesity and prostate cancer

The Journal of Clinical Endocrinology & Metabolism ◽

10.1210/clinem/dgab516 ◽

2021 ◽

Author(s):

Juan M Jiménez-Vacas ◽

Antonio J Montero-Hidalgo ◽

Enrique Gómez-Gómez ◽

Antonio C Fuentes-Fayos ◽

Francisco Ruiz-Pino ◽

...

Keyword(s):

Prostate Cancer ◽

High Specificity ◽

Metabolic Status ◽

Grey Zone ◽

Clinical Value ◽

Related Factors ◽

Clinical Model ◽

Non Invasive ◽

Splicing Variant ◽

Specificity And Sensitivity

Abstract Context Recent studies emphasize the importance of considering the metabolic status to develop personalized medicine approaches. This is especially relevant in prostate cancer (PCa), wherein the diagnostic capability of PSA dramatically drops when considering patients with PSA levels ranging 3-10 ng/mL, the so-called “grey-zone”. Hence, additional non-invasive diagnostic and/or prognostic PCa biomarkers are urgently needed, especially in the metabolic-status context. Objective To assess the potential relation of urine In1-ghrelin (a ghrelin splicing variant) levels with metabolic-related/pathological conditions (e.g. obesity/diabetes/BMI/insulin-glucose levels), and to define its potential clinical value in PCa (diagnostic/prognostic capacity) and relationship with PCa-risk in patients with PSA in the grey-zone. Methods Urine In1-ghrelin levels were measured by radioimmunoassay in a clinically/metabolically/pathologically well-characterized cohort of patients without (n=397) or with (n=213) PCa with PSA in the grey-zone. Results Key obesity-related factors associated with PCa-risk (BMI/diabetes/glucose/insulin) were strongly correlated to In1-ghrelin levels. Importantly, In1-ghrelin levels were higher in PCa patients compared to control patients (with suspect of PCa but negative-biopsy). Moreover, high In1-ghrelin levels were associated with increased PCa-risk and linked to PCa-aggressiveness (e.g. tumour-stage/lymphovascular-invasion). In1-ghrelin levels added significant diagnostic value to a clinical model consisting of age, suspicious-DRE, previous-biopsy, and PSA levels. Furthermore, a multivariate model consisting of clinical and metabolic variables, including In1-ghrelin levels, showed high specificity and sensitivity to diagnose PCa (AUC=0.740). Conclusions Urine In1-ghrelin levels are associated with obesity-related factors and PCa risk/aggressiveness, and could represent a novel and valuable non-invasive PCa biomarker, as well as a potential link in the pathophysiological relationship between obesity and PCa.

Download Full-text

Deep learning–assisted prostate cancer detection on bi-parametric MRI: minimum training data size requirements and effect of prior knowledge

European Radiology ◽

10.1007/s00330-021-08320-y ◽

2021 ◽

Author(s):

Matin Hosseinzadeh ◽

Anindo Saha ◽

Patrick Brand ◽

Ilse Slootweg ◽

Maarten de Rooij ◽

...

Keyword(s):

Prostate Cancer ◽

Deep Learning ◽

Prior Knowledge ◽

Training Data ◽

Expert Performance ◽

Free Response ◽

Algorithm Performance ◽

Prostate Imaging ◽

Key Points ◽

Clinically Significant

Abstract Objectives To assess Prostate Imaging Reporting and Data System (PI-RADS)–trained deep learning (DL) algorithm performance and to investigate the effect of data size and prior knowledge on the detection of clinically significant prostate cancer (csPCa) in biopsy-naïve men with a suspicion of PCa. Methods Multi-institution data included 2734 consecutive biopsy-naïve men with elevated PSA levels (≥ 3 ng/mL) that underwent multi-parametric MRI (mpMRI). mpMRI exams were prospectively reported using PI-RADS v2 by expert radiologists. A DL framework was designed and trained on center 1 data (n = 1952) to predict PI-RADS ≥ 4 (n = 1092) lesions from bi-parametric MRI (bpMRI). Experiments included varying the number of cases and the use of automatic zonal segmentation as a DL prior. Independent center 2 cases (n = 296) that included pathology outcome (systematic and MRI targeted biopsy) were used to compute performance for radiologists and DL. The performance of detecting PI-RADS 4–5 and Gleason > 6 lesions was assessed on 782 unseen cases (486 center 1, 296 center 2) using free-response ROC (FROC) and ROC analysis. Results The DL sensitivity for detecting PI-RADS ≥ 4 lesions was 87% (193/223, 95% CI: 82–91) at an average of 1 false positive (FP) per patient, and an AUC of 0.88 (95% CI: 0.84–0.91). The DL sensitivity for the detection of Gleason > 6 lesions was 85% (79/93, 95% CI: 77–83) @ 1 FP compared to 91% (85/93, 95% CI: 84–96) @ 0.3 FP for a consensus panel of expert radiologists. Data size and prior zonal knowledge significantly affected performance (4%, $$p<0.05$$ p < 0.05 ). Conclusion PI-RADS-trained DL can accurately detect and localize Gleason > 6 lesions. DL could reach expert performance using substantially more than 2000 training cases, and DL zonal segmentation. Key Points • AI for prostate MRI analysis depends strongly on data size and prior zonal knowledge. • AI needs substantially more than 2000 training cases to achieve expert performance.

Download Full-text

A shared decision support system for localized prostate cancer.

Journal of Clinical Oncology ◽

10.1200/jco.2012.30.34_suppl.326 ◽

2012 ◽

Vol 30 (34_suppl) ◽

pp. 326-326

Author(s):

Pavan Dadlani ◽

Jingyu Zhang ◽

Sebastian P. M. Dries ◽

Colleen M. Ennett ◽

Esther Toet

Keyword(s):

Prostate Cancer ◽

Decision Making ◽

Decision Support ◽

Decision Support System ◽

Support System ◽

Shared Decision ◽

Clinical Workflow ◽

User Research ◽

Ongoing Research ◽

Unwarranted Variation

326 Background: In the US, about 1 in 6 men is diagnosed with prostate cancer (PCa). Over 90% of them are localized PCa patients, which have the most controversial treatment decisions with few certainties about outcomes such as survival years and quality of life (QoL). Shared decision making is emerging, where patients need to make trade-offs between longevity and QoL based on personal preferences and treatment outcomes. Methods: By collaborating with leading PCa centers, medical psychologists and health economists, we investigate, iteratively design, and eventually test a decision support solution that could enhance treatment decision making and patient-clinician interaction. We interviewed 7 PCa clinicians and 13 patients and survivors, and observed 4 patient-clinician consultations. Results: Key insights from the user research: 1) existing decision aids are very generic and not personalized to the patient’s preferences, are not integrated in the clinical workflow, and involve a complex user experience; 2) there is a significant amount of unwarranted variation in PCa treatment (i.e. not preference sensitive, or patients lack the confidence to choose non-aggressive options that could lead to similar or better outcomes, e.g., active surveillance; and 3) clinicians need to understand what the patient’s preferences are (verbally discussed only), which consumes significant time in consultations. These insights led to developing a shared decision support system based on algorithms that use quantitative computations of quality adjusted life years (QALYs) and patient-friendly interactions. This system can be integrated in the clinical workflow, allow patients to make better informed decisions, and increase their confidence to choose the best treatment option according to their own preferences. We aim to increase patients’ involvement and satisfaction, enhancing consultation efficiency, and reducing unwarranted variation. Conclusions: Our ongoing research, motivated by user insights, focuses on developing shared decision support technology that is personalized to the patient’s profile and sensitive to their preferences. We will deploy validation studies at clinical sites and evaluate the system across the predefined outcome measures.

Download Full-text

Evaluation of MicroRNAs as Non-Invasive Diagnostic Markers in Urinary Cells from Patients with Suspected Prostate Cancer

Diagnostics ◽

10.3390/diagnostics10080578 ◽

2020 ◽

Vol 10 (8) ◽

pp. 578

Author(s):

Angelika Borkowetz ◽

Andrea Lohse-Fischer ◽

Jana Scholze ◽

Ulrike Lotzkat ◽

Christian Thomas ◽

...

Keyword(s):

Prostate Cancer ◽

Molecular Markers ◽

Diagnostic Performance ◽

Operating Characteristic ◽

Clinical Performance ◽

Diagnostic Value ◽

Non Invasive ◽

Diagnostic Potential ◽

Specificity And Sensitivity ◽

Diagnostic Power

Currently used tumor markers for early diagnosis of prostate cancer (PCa) are often lacking sufficient specificity and sensitivity. Therefore, the diagnostic potential of selected microRNAs in comparison to serum PSA levels and PSA density (PSAD) was explored. A panel of 12 PCa-associated microRNAs was quantified by qPCR in urinary sediments from 50 patients with suspected PCa undergoing prostate biopsy, whereupon PCa was detected in 26 patients. Receiver operating characteristic (ROC) curve analyses revealed a potential for non-invasive urine-based PCa detection for miR-16 (AUC = 0.744, p = 0.012; accuracy = 76%) and miR-195 (AUC = 0.729, p = 0.017; accuracy = 70%). While serum PSA showed an insufficient diagnostic value (AUC = 0.564, p = 0.656; accuracy = 50%) in the present cohort, PSAD displayed an adequate diagnostic performance (AUC = 0.708, p = 0.031; accuracy = 70%). Noteworthy, the combination of PSAD with the best candidates miR-16 and miR-195 either individually or simultaneously improved the diagnostic power (AUC = 0.801–0.849, p < 0.05; accuracy = 76–90%). In the sub-group of patients with PSA ≤ 10 ng/mL (n = 34), an inadequate diagnostic power of PSAD alone (AUC = 0.595, p = 0.524; accuracy = 68%) was markedly surpassed by miR-16 and miR-195 individually as well as by their combination with PSAD (AUC = 0.772–0.882, p < 0.05; accuracy = 74–85%). These findings further highlight the potential of urinary microRNAs as molecular markers with high clinical performance. Overall, these results need to be validated in a larger patient cohort.

Download Full-text

Assessment of miR-98-5p, miR-152-3p, miR-326 and miR-4289 Expression as Biomarker for Prostate Cancer Diagnosis

International Journal of Molecular Sciences ◽

10.3390/ijms20051154 ◽

2019 ◽

Vol 20 (5) ◽

pp. 1154 ◽

Cited By ~ 12

Author(s):

Leire Moya ◽

Jonelle Meijer ◽

Sarah Schubert ◽

Farhana Matin ◽

Jyotsna Batra

Keyword(s):

Prostate Cancer ◽

Cancer Progression ◽

Specific Antigen ◽

Area Under The Curve ◽

High Specificity ◽

Overall Survival Rate ◽

Non Invasive ◽

Specificity And Sensitivity ◽

Depth Analysis ◽

The Us

Prostate cancer (PCa) is one of the most commonly diagnosed cancers worldwide, accounting for almost 1 in 5 new cancer diagnoses in the US alone. The current non-invasive biomarker prostate specific antigen (PSA) has lately been presented with many limitations, such as low specificity and often associated with over-diagnosis. The dysregulation of miRNAs in cancer has been widely reported and it has often been shown to be specific, sensitive and stable, suggesting miRNAs could be a potential specific biomarker for the disease. Previously, we identified four miRNAs that are significantly upregulated in plasma from PCa patients when compared to healthy controls: miR-98-5p, miR-152-3p, miR-326 and miR-4289. This panel showed high specificity and sensitivity in detecting PCa (area under the curve (AUC) = 0.88). To investigate the specificity of these miRNAs as biomarkers for PCa, we undertook an in depth analysis on these miRNAs in cancer from the existing literature and data. Additionally, we explored their prognostic value found in the literature when available. Most studies showed these miRNAs are downregulated in cancer and this is often associated with cancer progression and poorer overall survival rate. These results suggest our four miRNA signatures could potentially become a specific PCa diagnostic tool of which prognostic potential should also be explored.

Download Full-text

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

Cancers ◽

10.3390/cancers12061606 ◽

2020 ◽

Vol 12 (6) ◽

pp. 1606

Author(s):

Jose M. Castillo T. ◽

Muhammad Arif ◽

Wiro J. Niessen ◽

Ivo G. Schoots ◽

Jifke F. Veenland

Keyword(s):

Machine Learning ◽

Decision Making ◽

Deep Learning ◽

Clinical Decision Making ◽

External Validation ◽

Clinical Decision ◽

Interquartile Range ◽

Added Value ◽

Receiver Operating Curve ◽

Learning Approaches

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.

Download Full-text