Prediction of Bone Marrow Biopsy Results From MRI in Multiple Myeloma Patients Using Deep Learning and Radiomics.

OBJECTIVES: In multiple myeloma and its precursor stages, plasma cell infiltration (PCI) and cytogenetic aberrations are important for staging, risk stratification, and response assessment. However, invasive bone marrow (BM) biopsies cannot be performed frequently and multifocally to assess the spatially heterogenous tumor tissue. Therefore, the goal of this study was to establish an automated framework to predict local BM biopsy results from magnetic resonance imaging (MRI). MATERIALS AND METHODS: This retrospective multicentric study used data from center 1 for algorithm training and internal testing, and data from center 2 to 8 for external testing. An nnU-Net was trained for automated segmentation of pelvic BM from T1-weighted whole-body MRI. Radiomics features were extracted from these segmentations, and random forest models were trained to predict PCI and the presence or absence of cytogenetic aberrations. Pearson correlation coefficient and the area under the receiver operating characteristic were used to evaluate the prediction performance for PCI and cytogenetic aberrations, respectively. RESULTS: A total of 672 MRIs from 512 patients (median age, 61 years; interquartile range, 53-67 years; 307 men) from 8 centers and 370 corresponding BM biopsies were included. The predicted PCI from the best model was significantly correlated ( P ≤ 0.01) to the actual PCI from biopsy in all internal and external test sets (internal test set: r = 0.71 [0.51, 0.83]; center 2, high-quality test set: r = 0.45 [0.12, 0.69]; center 2, other test set: r = 0.30 [0.07, 0.49]; multicenter test set: r = 0.57 [0.30, 0.76]). The areas under the receiver operating characteristic of the prediction models for the different cytogenetic aberrations ranged from 0.57 to 0.76 for the internal test set, but no model generalized well to all 3 external test sets. CONCLUSIONS: The automated image analysis framework established in this study allows for noninvasive prediction of a surrogate parameter for PCI, which is significantly correlated to the actual PCI from BM biopsy.

Imaging Artifacts of Nonadhesive Liquid Embolic Agents in Conventional and Cone-beam CT in a Novel in Vitro AVM Model.

BACKGROUND: A major drawback of liquid embolic agents (LEAs) is the generation of imaging artifacts (IA), which may represent a crucial obstacle for the detection of periprocedural hemorrhage or subsequent radiosurgery of cerebral arteriovenous malformations (AVMs). This study aimed to compare the IAs of Onyx, Squid and PHIL in a novel three-dimensional in vitro AVM model in conventional computed tomography (CT) and cone-beam CT (CBCT). METHODS: Tubes with different diameters were configured in a container resembling an AVM with an artificial nidus at its center. Subsequently, the AVM models were filled with Onyx 18, Squid 18, PHIL 25% or saline and inserted into an imaging phantom (n = 10/LEA). Afterwards CT and CBCT scans were acquired. The degree of IAs was graded quantitatively (Hounsfield units in a defined region of interest) and qualitatively (feasibility of defining the nidus)-Onyx vs. Squid vs. PHIL vs. saline, respectively. RESULTS: Quantitative density evaluation demonstrated more artifacts for Onyx compared to Squid and PHIL, e.g. 48.15 ± 14.32 HU for Onyx vs. 7.56 ± 1.34 HU for PHIL in CT (p < 0.001) and 41.88 ± 7.22 density units (DU) for Squid vs. 35.22 ± 5.84 DU for PHIL in CBCT (p = 0.044). Qualitative analysis showed less artifacts for PHIL compared to Onyx and Squid in both imaging modalities while there was no difference between Onyx and Squid regarding the definition of the nidus (p > 0.999). CONCLUSION: In this novel three-dimensional in vitro AVM model, IAs were higher for the EVOH/tantalum-based LEAs Onyx and Squid compared to iodine-based PHIL. Onyx induced the highest degree of IAs with only minor differences to Squid.

Biologically consistent dose accumulation using daily patient imaging.

BACKGROUND: This work addresses a basic inconsistency in the way dose is accumulated in radiotherapy when predicting the biological effect based on the linear quadratic model (LQM). To overcome this inconsistency, we introduce and evaluate the concept of the total biological dose, bEQDd. METHODS: Daily computed tomography imaging of nine patients treated for prostate carcinoma with intensity-modulated radiotherapy was used to compute the delivered deformed dose on the basis of deformable image registration (DIR). We compared conventional dose accumulation (DA) with the newly introduced bEQDd, a new method of accumulating biological dose that considers each fraction dose and tissue radiobiology. We investigated the impact of the applied fractionation scheme (conventional/hypofractionated), uncertainties induced by the DIR and by the assigned α/ß-value. RESULTS: bEQDd was systematically higher than the conventionally accumulated dose with difference hot spots of 3.3-4.9 Gy detected in six out of nine patients in regions of high dose gradient in the bladder and rectum. For hypofractionation, differences are up to 8.4 Gy. The difference amplitude was found to be in a similar range to worst-case uncertainties induced by DIR and was higher than that induced by α/ß. CONCLUSION: Using bEQDd for dose accumulation overcomes a potential systematic inaccuracy in biological effect prediction based on accumulated dose. Highest impact is found for serial-type late responding organs at risk in dose gradient regions and for hypofractionation. Although hot spot differences are in the order of several Gray, in dose-volume parameters there is little difference compared with using conventional or biological DA. However, when local dose information is used, e.g. dose surface maps, difference hot spots can potentially change outcomes of dose-response modelling and adaptive treatment strategies.

Impact of Single Dose Photons and Carbon Ions on Perfusion and Vascular Permeability: A Dynamic Contrast-Enhanced MRI Pilot Study in the Anaplastic Rat Prostate Tumor R3327-AT1.

We collected initial quantitative information on the effects of high-dose carbon (12C) ions compared to photons on vascular damage in anaplastic rat prostate tumors, with the goal of elucidating differences in response to high-LET radiation, using dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI). Syngeneic R3327-AT1 rat prostate tumors received a single dose of either 16 or 37 Gy 12C ions or 37 or 85 Gy 6 MV photons (iso-absorbed and iso-effective doses, respectively). The animals underwent DCE-MRI prior to, and on days 3, 7, 14 and 21 postirradiation. The extended Tofts model was used for pharmacokinetic analysis. At day 21, tumors were dissected and histologically examined. The results of this work showed the following: 1. 12C ions led to stronger vascular changes compared to photons, independent of dose; 2. Tumor growth was comparable for all radiation doses and modalities until day 21; 3. Nonirradiated, rapidly growing control tumors showed a decrease in all pharmacokinetic parameters (area under the curve, Ktrans, ve, vp) over time; 4. 12C-ion-irradiated tumors showed an earlier increase in area under the curve and Ktrans than photon-irradiated tumors; 5. 12C-ion irradiation resulted in more homogeneous parameter maps and histology compared to photons; and 6. 12C-ion irradiation led to an increased microvascular density and decreased proliferation activity in a largely dose-independent manner compared to photons. Postirradiation changes related to 12C ions and photons were detected using DCE-MRI, and correlated with histological parameters in an anaplastic experimental prostate tumor. In summary, this pilot study demonstrated that exposure to 12C ions increased the perfusion and/or permeability faster and led to larger changes in DCE-MRI parameters resulting in increased vessel density and presumably less hypoxia at the end of the observation period when compared to photons. Within this study no differences were found between curative and sub-curative doses in either modality.

Incorporation of Dosimetric Gradients and Parotid Gland Migration Into Xerostomia Prediction.

Purpose: Due to the sharp gradients of intensity-modulated radiotherapy (IMRT) dose distributions, treatment uncertainties may induce substantial deviations from the planned dose during irradiation. Here, we investigate if the planned mean dose to parotid glands in combination with the dose gradient and information about anatomical changes during the treatment improves xerostomia prediction in head and neck cancer patients. Materials and methods: Eighty eight patients were retrospectively analyzed. Three features of the contralateral parotid gland were studied in terms of their association with the outcome, i.e., grade ≥ 2 (G2) xerostomia between 6 months and 2 years after radiotherapy (RT): planned mean dose (MD), average lateral dose gradient (GRADX), and parotid gland migration toward medial (PGM). PGM was estimated using daily megavoltage computed tomography (MVCT) images. Three logistic regression models where analyzed: based on (1) MD only, (2) MD and GRADX, and (3) MD, GRADX, and PGM. Additionally, the cohort was stratified based on the median value of GRADX, and a univariate analysis was performed to study the association of the MD with the outcome for patients in low- and high-GRADX domains. Results: The planned MD failed to recognize G2 xerostomia patients (AUC = 0.57). By adding the information of GRADX (second model), the model performance increased to AUC = 0.72. The addition of PGM (third model) led to further improvement in the recognition of the outcome (AUC = 0.79). Remarkably, xerostomia patients in the low-GRADX domain were successfully identified (AUC = 0.88) by the MD alone. Conclusions: Our results indicate that GRADX and PGM, which together serve as a proxy of dosimetric changes, provide valuable information for xerostomia prediction.

In vivo assessment of cold stimulation effects on the fat fraction of brown adipose tissue using DIXON MRI.

PURPOSE: To evaluate the volume and changes of human brown adipose tissue (BAT) in vivo following exposure to cold using magnetic resonance imaging (MRI). MATERIALS AND METHODS: The clavicular region of 10 healthy volunteers was examined with a 3T MRI system. One volunteer participated twice. A cooling vest that was circulated with temperature-controlled water was used to expose each volunteer to a cold environment. Three different water temperature phases were employed: baseline (23°C, 20 min), cooling (12°C, 90 min), and a final warming phase (37°C, 30 min). Temperatures of the water in the circuit, of the body, and at the back skin of the volunteers were monitored with fiberoptic temperature probes. Applying the 2-point DIXON pulse sequence every 5 minutes, fat fraction (FF) maps were determined and evaluated over time to distinguish between brown and white adipose tissue. RESULTS: Temperature measurements showed a decrease of 3.8 ± 1.0°C of the back skin temperature, while the body temperature stayed constant at 37.2 ± 0.9°C. Focusing on the two interscapular BAT depots, a mean FF decrease of -2.9 ± 2.0%/h (P < 0.001) was detected during cold stimulation in a mean absolute volume of 1.31 ± 1.43 ml. Also, a correlation of FF decrease to back skin temperature decrease was observed in all volunteers (correlation coefficients: |r| = [0.51; 0.99]). CONCLUSION: We found that FF decreases in BAT begin immediately with mild cooling of the body and continue during long-time cooling. LEVEL OF EVIDENCE: 2 J. Magn. Reson. Imaging 2017;45:369-380.

Cortical vessel sign on susceptibility weighted imaging reveals clinically relevant hypoperfusion in internal carotid artery stenosis.

BACKGROUND AND PURPOSE: Internal carotid artery (ICA) stenosis can lead to cerebral hypoperfusion and is a common cause of stroke. As susceptibility weighted imaging (SWI) has been used for penumbra imaging in acute ischemic stroke, we aimed at analyzing hypoperfusion using SWI in patients with ICA stenosis. MATERIAL AND METHODS: Clinical characteristics, asymmetric cortical vessel sign (more and/or larger, hypointense asymmetric cortical vessels) on SWI, Doppler sonography results and diffusion weighted imaging (DWI) lesion volume were retrospectively analyzed in patients with ICA stenosis. In a subgroup of patients, volume of prolonged time to peak and volume of prolonged time to peak of the residue curve (Tmax) were measured as reference standard. Outcome was assessed as modified Rankin score at discharge. RESULTS: 104 patients were included. Median age was 72 and median degree of stenosis 70% according to NASCET. 13% had a asymmetric cortical vessel sign. These patients had a higher degree of stenosis (80% vs. 70%, p=0.004), were more often symptomatic (93% vs. 61%, p=0.020) and had higher DWI volume (7.3ml vs. 0.2ml, p=0.011). Specificity for the prediction of DWI lesions was 86%. Also, patients with asymmetric cortical vessel sign had lower rates of favorable outcome (mRS=0-2; 57% vs. 82%, p=0.033) and volumes of Tmax≥4s, ≥6s, ≥8s, ≥10s and TTP≥2s, ≥4s, ≥6s were significantly higher. In multivariate analysis, asymmetric cortical vessel sign was an independent negative predictor of favorable outcome (mRS 0-2; OR 0.184; CI [0.039; 0.875] p=0.033). CONCLUSION: In patients with ICA stenosis, asymmetric cortical vessel sign is a sign of clinically relevant hypoperfusion.

Prognostic value of combined visualization of MR diffusion and perfusion maps in glioblastoma.

We analyzed whether the combined visualization of decreased apparent diffusion coefficient (ADC) values and increased cerebral blood volume (CBV) in perfusion imaging can identify prognosis-related growth patterns in patients with newly diagnosed glioblastoma. Sixty-five consecutive patients were examined with diffusion and dynamic susceptibility-weighted contrast-enhanced perfusion weighted MRI. ADC and CBV maps were co-registered on the T1-w image and a region of interest (ROI) was manually delineated encompassing the enhancing lesion. Within this ROI pixels with ADC values the 70th percentile (CBVmax) and the intersection of pixels with ADCmin and CBVmax were automatically calculated and visualized. Initially, all tumors with a mean intersection greater than the upper quartile of the normally distributed mean intersection of all patients were subsumed to the first growth pattern termed big intersection (BI). Subsequently, the remaining tumors' growth patterns were categorized depending on the qualitative representation of ADCmin, CBVmax and their intersection. Log-rank test exposed a significantly longer overall survival of BI (n = 16) compared to non-BI group (n = 49) (p = 0.0057). Thirty-one, four and 14 patients of the non-BI group were classified as predominant ADC-, CBV- and mixed growth group, respectively. In a multivariate Cox regression model, the BI-, CBV- and mixed groups had significantly lower adjusted hazard ratios (p-value, α(Bonferroni) < 0.006) when compared to the reference group ADC: 0.29 (0.0027), 0.11 (0.038) and 0.33 (0.0059). Our study provides evidence that the combination of diffusion and perfusion imaging allows visualization of different glioblastoma growth patterns that are associated with prognosis. A possible biological hypothesis for this finding could be the interpretation of the ADCmin fraction as the invasion-front of tumor cells while the CBVmax fraction might represent the vascular rich tumor border that is "trailing behind" the invasion-front in the ADC group.

Software for quantitative analysis of radiotherapy: overview, requirement analysis and design solutions.

Radiotherapy is a fast-developing discipline which plays a major role in cancer care. Quantitative analysis of radiotherapy data can improve the success of the treatment and support the prediction of outcome. In this paper, we first identify functional, conceptional and general requirements on a software system for quantitative analysis of radiotherapy. Further we present an overview of existing radiotherapy analysis software tools and check them against the stated requirements. As none of them could meet all of the demands presented herein, we analyzed possible conceptional problems and present software design solutions and recommendations to meet the stated requirements (e.g. algorithmic decoupling via dose iterator pattern; analysis database design). As a proof of concept we developed a software library "RTToolbox" following the presented design principles. The RTToolbox is available as open source library and has already been tested in a larger-scale software system for different use cases. These examples demonstrate the benefit of the presented design principles.

Development and validation of automatic tools for interactive recurrence analysis in radiation therapy: optimization of treatment algorithms for locally advanced pancreatic cancer.

BACKGROUND: In radiation oncology recurrence analysis is an important part in the evaluation process and clinical quality assurance of treatment concepts. With the example of 9 patients with locally advanced pancreatic cancer we developed and validated interactive analysis tools to support the evaluation workflow. METHODS: After an automatic registration of the radiation planning CTs with the follow-up images, the recurrence volumes are segmented manually. Based on these volumes the DVH (dose volume histogram) statistic is calculated, followed by the determination of the dose applied to the region of recurrence and the distance between the boost and recurrence volume. We calculated the percentage of the recurrence volume within the 80%-isodose volume and compared it to the location of the recurrence within the boost volume, boost + 1 cm, boost + 1.5 cm and boost + 2 cm volumes. RESULTS: Recurrence analysis of 9 patients demonstrated that all recurrences except one occurred within the defined GTV/boost volume; one recurrence developed beyond the field border/outfield. With the defined distance volumes in relation to the recurrences, we could show that 7 recurrent lesions were within the 2 cm radius of the primary tumor. Two large recurrences extended beyond the 2 cm, however, this might be due to very rapid growth and/or late detection of the tumor progression. CONCLUSION: The main goal of using automatic analysis tools is to reduce time and effort conducting clinical analyses. We showed a first approach and use of a semi-automated workflow for recurrence analysis, which will be continuously optimized. In conclusion, despite the limitations of the automatic calculations we contributed to in-house optimization of subsequent study concepts based on an improved and validated target volume definition.

A method to visualize the uncertainty of the prediction of radiobiological models.

A method for quantitative visualization of the uncertainty in the predicted tumor control probability (TCP) and normal tissue complication probability (NTCP) in radiotherapy has been developed. Uncertainties of TCP and NTCP due to inter-individual variation of the underlying radiosensitivity parameters was simulated by sampling the prescribed dose from a uniform distribution and the radiosensitivity-parameters from a Gaussian distribution. The result is visualized as a scatter-plot superimposed to the population-based dose response curves using the prescribed dose as the common dosimetric variable. In addition, probability histograms are derived quantifying the probability of specific TCP- or NTCP-values for individual patients from the underlying population. The method is exemplified with a pleural mesothelioma case with the lung as organ at risk. A prescribed dose of 54 Gy together with radiosensitivity variations of 6% (tumor) and 10% (normal tissue) results in a TCP of 85% (range 68-94%, 90% confidence interval, CI) and an NTCP of 4% (range 3-6%, 90% CI), respectively. Increasing the radiosensitivity variation of the tumor to 15% and reducing the lung tolerance dose by 25% results in values of 84% (range 51-97%, 90% CI) for TCP and 9% (range 6-12%, 90% CI) for NTCP. Increasing the dose to 60 Gy leads to TCP- and NTCP-values of 93% (range 69-100%, 90% CI) and 12% (range 8-17%, 90% CI), respectively. The new method visualizes the uncertainty of TCP- and NTCP-values and hence of the therapeutic window. This can help the clinician to assess the treatment plan for the individual patient.