RPS 116 - Oncologic imaging in genitourinary: kidney and prostate, advanced topics

Purpose:

To analyse the performance of different prostate imaging features associated with prostate cancer (PCa) extracapsular extension (ECE) in order to develop an ECE-score.

Methods and materials:

A retrospective study on 114 patients who underwent multiparameter prostate MRI (mp-MRI) at 1.5Tesla using a 32 phased-array coil before radical prostatectomy was conducted. mp-MRIs were analysed for the length of contact between PCa and the prostatic capsule, capsule irregularity, bulge, loss of capsule, neurovascular bundle thickening, measurable extra-capsular disease, and DWI signal increase in extra-capsular location, and ADC value and PSA density were calculated. The data was analysed with a parametric test and 5 diagnostic parameters. A threshold of 90% in specificity and positive predictive value was used to distinguish between major and minor criteria. The ECE-score was tested with the ROC curve procedure.

Results:

ECE was found at histopathological review in 41 patients (ECE+). Capsule irregularity, bulge, loss of capsule, neurovascular bundle thickening, DWI signal increase in extra-capsular location, and measurable extra-capsular disease had a highly significant difference between ECE+ and others (p<0.0001). Capsule irregularity, bulge, and loss of capsule were included in the minor criteria and neurovascular bundle thickening and DWI signal increase in the extra-capsular location were included in the major criteria. The ECE-score was equal to 1 for the presence of 0 criteria, 2 for 1 minor, 3 for 2 minor, 4 for 3 minor or 1 major, and 5 for 2 major criteria or the presence of measurable extra-capsular disease. The ROC curve procedure established the good ability of the ECE-score to discriminate ECE+ with an AUC of 0.93 (0.87-0.98).

Conclusion:

The ECE-score proposed can improve the detection of ECE and allow for more accurate staging, providing important additional information for optimal patient-tailored treatment planning.

Limitations:

A monocentric and retrospective study. The absence of interobserver variability.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To investigate the influence of the minimum b-value in diffusion parameters estimated for prostate tissues using mono-exponential and kurtosis models.

Methods and materials:

58 patients were prospectively enrolled to perform magnetic resonance imaging of the prostate at 3.0T. A diffusion-weighted sequence using 11 b-values ranging from 0-2,000 s/mm² was used.

Mono-exponential and kurtosis models were fitted to the diffusion signal. 6 different b-value combinations were used to estimate the apparent diffusion coefficient (ADC) while varying the minimum b-value (0, 50, 100, 150, 200, and 500 s/mm²) and b-values up to 1,100 s/mm². Diffusion kurtosis imaging (DKI) metrics were computed using b-values ranging from 0-2,000 s/mm² and considering the same minimum b-values.

The Mann-Whitney test was used to assess differences in diffusion metrics between prostate cancer (PCa) and normal tissue. Differences in b-value combinations were assessed with the Friedman test. Receiver operating characteristics curves were performed for each metric.

Results:

ADC and the mean diffusivity (MD) were significantly lower and the mean kurtosis (MK) was significantly higher (p<0.001) in PCa compared to normal tissue for all b-values combinations.

MK did not change with the minimum b-value for normal tissue or PCa (p>0.05), but ADC and MD differed significantly.

The MK metric achieved the highest AUC (96.0%) and accuracy (93.8%) using all b-values (0-2,000 s/mm²) but the diagnostic performance was not statistically improved when compared to ADC or MD.

Conclusion:

MK values were not influenced by the chosen low b-value, indicating that microperfusion effects do not influence MK, contrarily to ADC and MD.

The diagnostic performances of ADC, MD, and MK were similar.

Limitations:

Prostate lesions were confirmed by transrectal ultrasound-guided biopsy.

Ethics committee approval

Approved by the hospital ethics committee [number 251/12 (190-DEFI/195-CES)] and all patients gave written informed consent.

Funding:

No funding was received for this work.

Purpose:

Artificial intelligence (AI) offers new possibilities for obtaining objective quantitative measurements of biomarkers from positron-emission tomography/computed tomography (PET/CT). The aim of this study was to evaluate the prognostic significance of automated quantification of tumour infiltration in the prostate gland in ¹⁸F-Choline PET/CT of high-risk prostate cancer patients in respect to age, PSA, and a Gleason score.

Methods and materials:

Training of the AI-algorithm was performed on PET/CT scans of 143 patients. Validation of the algorithm was carried out in ¹⁸F-choline PET/CT scans of 272 high-risk prostate cancer patients. Automated measurements of the radiotracer uptake in voxels in the region automatically segmented as prostate by the AI-algorithm were performed. Voxels with a standardised uptake value (SUV) above 2.65g/ml were considered abnormal. The following calculations were obtained: lesion VOLUME (volume of abnormal voxels, ml), total lesion uptake (TLU, defined as the product of SUVmean x lesion VOLUME), and lesion FRACTION (the quotient of VOLUME and total prostate volume). Associations between automated measurements, age, prostate-specific antigen (PSA) level (logarithmic), Gleason score, and prostate cancer-specific survival were studied using a univariate Cox-regression model.

Results:

20 patients died of prostate cancer during follow-up. The PET/CT measurements were significantly associated with prostate cancer-specific survival [VOLUME (HR 1.03, p <0.001); TLU (HR 1.9, p<0.001); lesion FRACTION (HR 1.05, p <0.001)], while age, PSA, and Gleason score were not.

Conclusion:

¹⁸F-choline uptake, determined by automatically-derived measurements, was significantly associated with disease-specific survival in this patient cohort.

Limitations:

The small number of events during the follow-up period in this patient group prevented us from performing multivariate analyses.

Ethics committee approval

Research ethical review board in Denmark and Sweden.

Funding:

Gothenburg University (Medical Faculty ALF Grants), Sweden, and EXINI Diagnostics AB, Lund, Sweden.

Purpose:

To discriminate between patients with high-risk (HR) and low-risk (LR) prostate cancer (PCa) in order to support radiologists in deciding on the most proper therapy strategy.

Methods and materials:

42 patients with a clinical suspicion of PCa were consecutively selected from the database of our institution. All patients underwent 3T-mpMRI and TRUS biopsy and, based on their Gleason scores (GS), were assigned to the HR (GS≥3+4) or LR class, the latter including, besides patients with GS=3+3, patients with a negative biopsy, whether they have positive or negative mpMRI. 84 radiomic features were extracted on DWI sequences and related ROC curves computed. The feature showing the lowest p-value in discriminating HR from LR was selected.

Results:

The mean of the local coefficient of variation (CV_L-m), representing local DWI variance, performed the best (p~10^-6) and discriminated HR from LR with AUC=0.91 (95% CI, 0.75-0.97), specificity=85%, sensitivity=87% (4 FP and 2 FN), and all FPs were GS=3+3. These results yielded the probability of FDR=24% of the overtreatment for LR patients and the probability of FOR=8% that a HR patient is not treated.

Conclusion:

One of our radiomic features derived from DWI sequences was enough to differentiate HR from LR PCa. Since the level of restriction to the motion of water molecules in the extracellular compartment affects tumour behaviour, radiomic features extracted from DWI sequences result in the best candidate to quantify relevant properties of tumour habitats needed to characterise the different tumour heterogeneities.

Limitations:

Patients were not enough to reliably include clinical parameters in the PCa risk assessment. Although they could be crucial to help to improve the radiomic model, a higher number of parameters require a number of patients growing exponentially to have a representative sample size.

Ethics committee approval

IRB approval. Written, informed consent was waived.

Funding:

No funding was received for this work.

Purpose:

To determine if a negative PSMA PET CT can ever rule out tumour recurrences in prostate cancer.

Methods and materials:

In a period of 6 months, 246 PET CT PSMAs with biochemical relapse were performed. Those with negative or indeterminate findings were selected to be studied with a multiparametric prostate MRI. 12 patients, 58-75 years old, with a PSA between 0.9 and 22 ng/ml were included.

Prostate multiparametric resonance was performed with a Discovery 750W General Electric 3 Tesla. The study protocol included axial T2 panoramic, axial and coronal T2 high resolution of the prostate, diffusion (DWI), and ADC focus and perfusion sequences.

PET CT acquisition was performed 60 minutes after intravenous administration of 2 MBq/Kg of 68Ga-PSMA with 64-slice equipment (General Electric Discovery 690 VCT) from the skull to mid-thigh. The images were corrected for “flight time” (TOF correction).

Results:

Of the 12 included patients, 10 presented with local tumour recurrences at the level of the prostate, periprostatic region, or bladder wall. These lesions were not seen with PET PSMA due to the concentration of the radiopharmaceutical in the bladder.

All lesions presented a pathological signal in T2, restriction in DWI, and early enhancement in the perfusion sequence. This last sequence is essential for the detection of tumour recurrences due to the neoangiogenesis of tumoural tissue.

Conclusion:

We can conclude that in cases of a negative PSMA PET CT, a multiparametric prostate MRI must be suggested. This study may show relapses in regions not seen with PET PSMA and may change patient management. The main sequence that represents the key of diagnosis is perfussion.

Limitations:

Few patients.

Ethics committee approval

Written informed consent obtained.

Funding:

No funding was received for this work.

Purpose:

To determine whether the quantitative fractional intracellular volume (FIC) from VERDICT MRI (vascular, extracellular, and restricted diffusion for cytometry in tumours) and/or ADC (apparent diffusion coefficient) can prospectively identify men undergoing prostate mpMRI with significant cancer.

Methods and materials:

We previously demonstrated FIC has a higher ROC-AUC (0.93) than ADC (0.85) for differentiating clinically-significant from benign/non-significant prostate cancer. In this study, we derived and prospectively applied FIC and ADC thresholds based on Youden’s index (from previous ROC-analysis using men with Likert≥3): FIC:0.41, ADC:1.12x10^-3 to a cohort of 30 men with Likert≥3 mpMRI lesions who underwent targeted biopsy.

The mean lesion FIC, ADC, sensitivity, and specificity of the derived thresholds were calculated.

Results:

Biopsies revealed 16 clinically-significant (3+4=2, ≥4+3=4) cancers and 14 benign/non-significant cancers (benign=12, 3+3=2). Clinically-significant lesions had higher FIC (mean: 0.43±0.22) compared to benign/non-significant lesions (mean: 0.26±0.15) p=0.035. ADC was not significantly different between the two groups (significant: 1.06 ±0.16 x10^-3 vs benign/non-significant: 0.93±0.33 x10^-3, p=0.12).

The FIC threshold correctly classified 86% of men (n=12) with a benign/non-significant biopsy as negative for significant-cancer compared with 36% of men (n=5) for the ADC threshold. 62% of men with significant cancer (n=10) were classified as positive for significant cancer by the FIC threshold, compared with 81% of men (n=13) using the ADC threshold.

Combining Likert≥4 and FIC threshold, 12/14 men with benign/non-significant pathology could have avoided biopsy. However, 1/16 men with significant cancer would not have been biopsied. Combining Likert≥4 and ADC threshold, 4/14 men with benign/non-significant pathology could have avoided a biopsy and all men with significant cancer would have undergone a biopsy.

Conclusion:

A combined Likert score and FIC thresholds could help avoid unnecessary biopsies in men being investigated for prostate cancer.

Limitations:

n/a

Ethics committee approval

London–Surrey Borders REC approval, written informed consent obtained.

Funding:

Prostate Cancer UK (PG14-018-TR2).

Purpose:

A preoperative distinction between renal oncocytoma (RON) and chromophobe renal cell carcinoma (cRCC) remains challenging based on the visual interpretation of multiparametric magnetic resonance imaging (mp-MRI) including apparent diffusion coefficient (ADC) or dynamic sequences. We aimed to evaluate the ability of radiomics, a recently emerged tool for mathematical tissue characterisation based on tumour heterogeneity, in differentiating both tumours.

Methods and materials:

This single-centre retrospective study included 14 patients with histopathologically proven RON (n=6) and cRCC (n=8). All cases were imaged before surgery by mp-MRI. Axial MRI-ADC mapping images were used for the manual segmentation of the masses by two radiologists using open-source PyRadiomics software. The radiomics features were extracted from 3 categories: shape and size, histogram-based first-order texture, and high-order texture. Interobserver reliability was assessed using the intraclass correlation coefficient (ICC). Features with excellent (ICC>0.90) agreement were compared between RON and cRCC groups. The features with a p<0.05 were analysed by the receiver operating curve for the precision of RON.

Results:

16 radiomics features (2 shape and size, 5 first-order, and 8 high-order texture) provided an ICC>0.90 and p<0.05. Histogram-based first-order features using 3-dimensional whole tumour ADC showed the highest diagnostic performance for predicting RON (median ADC cut-off ≥1.84x10⁻³ mm/s, sensitivity=100%, and specificity 100%, p=0.002). Among high order texture features, 5 grey-level run-length matrixes and 3 grey-level size zone matrixes related features predicted RON with considerably high performance (area under the curve >0.90, sensitivity and specificity >0.80, p<0.05).

Conclusion:

Radiomics analysis of MRI-ADC mapping can successfully distinguish between RON and cRCC with a quite high diagnostic precision performance.

Limitations:

The small sample size and retrospective design.

Ethics committee approval

Institutional ethics committee approval number: 2019/0283.

Funding:

No funding was received for this work.

Purpose:

Renal cancer is phenotypically and genetically highly heterogeneous. Genetic intratumoral heterogeneity plays a determining role in tumour aggressiveness, metastasis, and treatment resistance. Precise spatial registration of imaging and molecular data is required for multi-omics data integration to leverage the full potential of personalised cancer care. We describe the development of an automated pipeline for producing 3D-printed tumour moulds for image-guided tissue sampling.

Methods and materials:

Following preoperative multiparametric, morphologic, and physiological MRI and image registration, the renal tumour, normal kidney, and anatomical structures that needed preserving during tissue sampling were manually segmented. 3D-printed tumour moulds were generated and cutting-guides with 1 cm spacing were introduced. The direction of the cutting-guides can be selected according to pathologists' preferences. The design of the mould was improved in an iterative process involving pathologists, urologists, radiologists, and imaging scientists to ensure accuracy, universal applicability, and minimal disturbance to workflows. The accuracy of the spatial registration was determined by the Dice similarity coefficient between predicted and observed tissue cross-sections.

Results:

6 patients (age 62.7±8.3 years) were included in the iterative mould design process. High morphological correspondence of anatomy between tissue and predicted cross-sections from imaging were seen, as well as a good agreement of predicted and observed tumour cross-sections. In the latest iteration of the mould, Dice similarity coefficients of 0.92 and 0.76 were observed for the tumour and normal kidney, respectively, thanks to the introduction of orientation landmarks.

Conclusion:

A highly automated pipeline for the acquisition of tissue samples registered to imaging data was developed in an iterative process. The involvement of urologists and pathologists ensured maximum clinical acceptability of the mould and resulted in a pathway for multi-regional, multi-omics data integration.

Limitations:

Recruitement for this study is ongoing.

Ethics committee approval

UK-REC-Numbers: 5/EE/0378, 03/018.

Funding:

Cancer Research UK.

Purpose:

Radiomic analysis and quantification of tumour heterogeneity may capture additional treatment effects beyond current standards and predict clinical outcomes, however, how CT acquisition affects quantifiable parameters should be considered. This prospective study aims to explore the impact of an arterial or portal venous phase acquisition on radiomic features in patients with metastatic renal cell carcinoma.

Methods and materials:

33 adults (31 male; mean age 65.5±9.1) with 61 lesions underwent contrast-enhanced CT. Volumetric regions-of-interest were delineated around the primary±metastatic lesion(s) in both arterial and portal phases. Radiomic features were extracted using validated in-house software. A total of 183 first, second, and high-order statistical parameters were compared using a Wilcoxon signed-rank test, with a Bonferroni correction for multiple comparisons, p<0.005.

Results:

Pixel intensity parameters differed between phases e.g. the mean intensity (58.6 vs 68.6, p=0.0015), variance (4,778.9 vs 5,251.3, p=0.0019) (portovenous and arterial, respectively). There was no significant difference in shape, histogram, fractal, second-order grey-level co-occurrence matrix or high-order grey-level run length, grey-level dependence matrix, grey-level run-length size zone matrix, and neighbourhood grey-tone difference matrix parameters.

Conclusion:

As expected, histogram parameters differed between phases. However, the stability of radiomics parameters between phases indicates their robustness to acquisition timing and is promising for their use as biomarkers of the treatment response.

Limitations:

The small sample size and heterogeneity in CT acquisition parameters.

Ethics committee approval

Ethical approval and informed consent - NRES Committee North West Liverpool Central, 06/06/2011, REC ref: 11/NW/0246.

Funding:

The National Institute for Health Research Health Technology Assessment Programme (project number 09/91/21).

Purpose:

Immune checkpoint inhibitors (ICIs) now have a place in the treatment of metastatic renal cell carcinomas (mRCC). Atypical responses have been described under ICI in several types of cancer such as pseudoprogression that are not captured by the conventional RECIST 1.1 criteria. Therefore, new criteria have been proposed, irRC, irRECIST, and iRECIST, but little is known about their usefulness in mRCC evaluation.

Methods and materials:

This monocentric study included patients treated in a clinical trial with nivolumab for mRCC. We retrospectively analysed the response patterns, their frequency, the discrepancies between the RECIST 1.1, irRECIST, and iRECIST criteria, and overall survival by response type.

Results:

Of the 56 patients included, 81% presented a conventional response (objective response or stability 18% and progression 63%) and 19% an atypical response (pseudoprogression 5% and dissociated response 14%). The irRECIST and iRECIST criteria captured the phenomenon of pseudoprogression in all patients, but 3 had no therapeutic impact. Patients with a dissociated response had the lowest overall survival with a significant difference compared to the progression subgroup.

Conclusion:

Atypical patterns of response occur in 19% of the patients treated by nivolumab for mRCC. Further studies are needed to confirm our findings of dissociated response patients who seem to have a worse prognosis than classical progressive patients.

Limitations:

A monocentric study involving a small number of patients, which does not allow the results to be generalised and significant prognostic factors to be identified. The patients were included in a therapeutic protocol and this could have constituted a selection bias.

Ethics committee approval

n/a

Funding:

No funding was received for this work.