Accuracy of gross tumour volume delineation with [68Ga]-PSMA-PET compared to histopathology for high-risk prostate cancer.

BACKGROUND: The delineation of intraprostatic lesions is vital for correct delivery of focal radiotherapy boost in patients with prostate cancer (PC). Errors in the delineation could translate into reduced tumour control and potentially increase the side effects. The purpose of this study is to compare PET-based delineation methods with histopathology. MATERIALS AND METHODS: The study population consisted of 15 patients with confirmed high-risk PC intended for prostatectomy. [68Ga]-PSMA-PET/MR was performed prior to surgery. Prostate lesions identified in histopathology were transferred to the in vivo [68Ga]-PSMA-PET/MR coordinate system. Four radiation oncologists manually delineated intraprostatic lesions based on PET data. Various semi-automatic segmentation methods were employed, including absolute and relative thresholds, adaptive threshold, and multi-level Otsu threshold. RESULTS: The gross tumour volumes (GTVs) delineated by the oncologists showed a moderate level of interobserver agreement with Dice similarity coefficient (DSC) of 0.68. In comparison with histopathology, manual delineations exhibited the highest median DSC and the lowest false discovery rate (FDR) among all approaches. Among semi-automatic approaches, GTVs generated using standardized uptake value (SUV) thresholds above 4 (SUV > 4) demonstrated the highest median DSC (0.41), with 0.51 median lesion coverage ratio, FDR of 0.66 and the 95th percentile of the Hausdorff distance (HD95%) of 8.22 mm. INTERPRETATION: Manual delineations showed a moderate level of interobserver agreement. Compared to histopathology, manual delineations and SUV > 4 exhibited the highest DSC and the lowest HD95% values. The methods that resulted in a high lesion coverage were associated with a large overestimation of the size of the lesions.

Improving MR image quality with a multi-task model, using convolutional losses.

PURPOSE: During the acquisition of MRI data, patient-, sequence-, or hardware-related factors can introduce artefacts that degrade image quality. Four of the most significant tasks for improving MRI image quality have been bias field correction, super-resolution, motion-, and noise correction. Machine learning has achieved outstanding results in improving MR image quality for these tasks individually, yet multi-task methods are rarely explored. METHODS: In this study, we developed a model to simultaneously correct for all four aforementioned artefacts using multi-task learning. Two different datasets were collected, one consisting of brain scans while the other pelvic scans, which were used to train separate models, implementing their corresponding artefact augmentations. Additionally, we explored a novel loss function that does not only aim to reconstruct the individual pixel values, but also the image gradients, to produce sharper, more realistic results. The difference between the evaluated methods was tested for significance using a Friedman test of equivalence followed by a Nemenyi post-hoc test. RESULTS: Our proposed model generally outperformed other commonly-used correction methods for individual artefacts, consistently achieving equal or superior results in at least one of the evaluation metrics. For images with multiple simultaneous artefacts, we show that the performance of using a combination of models, trained to correct individual artefacts depends heavily on the order that they were applied. This is not an issue for our proposed multi-task model. The model trained using our novel convolutional loss function always outperformed the model trained with a mean squared error loss, when evaluated using Visual Information Fidelity, a quality metric connected to perceptual quality. CONCLUSION: We trained two models for multi-task MRI artefact correction of brain, and pelvic scans. We used a novel loss function that significantly improves the image quality of the outputs over using mean squared error. The approach performs well on real world data, and it provides insight into which artefacts it detects and corrects for. Our proposed model and source code were made publicly available.

Robust treatment planning of dose painting for prostate cancer based on ADC-to-Gleason score mappings - what is the potential to increase the tumor control probability?

BACKGROUND AND PURPOSE: The aim of this study was to evaluate the potential to increase the tumor control probability (TCP) with 'dose painting by numbers' (DPBN) plans optimized in a treatment planning system (TPS) compared to uniform dose plans. The DPBN optimization was based on our earlier published formalism for prostate cancer that is driven by dose-responses of Gleason scores mapped from apparent diffusion coefficients (ADC). MATERIAL AND METHODS: For 17 included patients, a set of DPBN plans were optimized in a TPS by maximizing the TCP for an equal average dose to the prostate volume (CTVT) as for a conventional uniform dose treatment. For the plan optimizations we applied different photon energies, different precisions for the ADC-to-Gleason mappings, and different CTVT positioning uncertainties. The TCP increasing potential was evaluated by the DPBN efficiency, defined as the ratio of TCP increases for DPBN plans by TCP increases for ideal DPBN prescriptions (optimized without considering radiation transport phenomena, uncertainties of the CTVT positioning, and uncertainties of the ADC-to-Gleason mapping). RESULTS: The median DPBN efficiency for the most conservative planning scenario optimized with a low precision ADC-to-Gleason mapping, and a positioning uncertainty of 0.6 cm was 10%, meaning that more than half of the patients had a TCP gain of at least 10% of the TCP for an ideal DPBN prescription. By increasing the precision of the ADC-to-Gleason mapping, and decreasing the positioning uncertainty the median DPBN efficiency increased by up to 40%. CONCLUSIONS: TCP increases with DPBN plans optimized in a TPS were found more likely with a high precision mapping of image data into dose-responses and a high certainty of the tumor positioning. These findings motivate further development to ensure precise mappings of image data into dose-responses and to ensure a high spatial certainty of the tumor positioning when implementing DPBN clinically.

Deep learning-based classification and structure name standardization for organ at risk and target delineations in prostate cancer radiotherapy.

Radiotherapy (RT) datasets can suffer from variations in annotation of organ at risk (OAR) and target structures. Annotation standards exist, but their description for prostate targets is limited. This restricts the use of such data for supervised machine learning purposes as it requires properly annotated data. The aim of this work was to develop a modality independent deep learning (DL) model for automatic classification and annotation of prostate RT DICOM structures. Delineated prostate organs at risk (OAR), support- and target structures (gross tumor volume [GTV]/clinical target volume [CTV]/planning target volume [PTV]), along with or without separate vesicles and/or lymph nodes, were extracted as binary masks from 1854 patients. An image modality independent 2D InceptionResNetV2 classification network was trained with varying amounts of training data using four image input channels. Channel 1-3 consisted of orthogonal 2D projections from each individual binary structure. The fourth channel contained a summation of the other available binary structure masks. Structure classification performance was assessed in independent CT (n = 200 pat) and magnetic resonance imaging (MRI) (n = 40 pat) test datasets and an external CT (n = 99 pat) dataset from another clinic. A weighted classification accuracy of 99.4% was achieved during training. The unweighted classification accuracy and the weighted average F1 score among different structures in the CT test dataset were 98.8% and 98.4% and 98.6% and 98.5% for the MRI test dataset, respectively. The external CT dataset yielded the corresponding results 98.4% and 98.7% when analyzed for trained structures only, and results from the full dataset yielded 79.6% and 75.2%. Most misclassifications in the external CT dataset occurred due to multiple CTVs and PTVs being fused together, which was not included in the training data. Our proposed DL-based method for automated renaming and standardization of prostate radiotherapy annotations shows great potential. Clinic specific contouring standards however need to be represented in the training data for successful use. Source code is available at https://github.com/jamtheim/DicomRTStructRenamerPublic.

Zero TE-based pseudo-CT image conversion in the head and its application in PET/MR attenuation correction and MR-guided radiation therapy planning.

PURPOSE: To describe a method for converting Zero TE (ZTE) MR images into X-ray attenuation information in the form of pseudo-CT images and demonstrate its performance for (1) attenuation correction (AC) in PET/MR and (2) dose planning in MR-guided radiation therapy planning (RTP). METHODS: Proton density-weighted ZTE images were acquired as input for MR-based pseudo-CT conversion, providing (1) efficient capture of short-lived bone signals, (2) flat soft-tissue contrast, and (3) fast and robust 3D MR imaging. After bias correction and normalization, the images were segmented into bone, soft-tissue, and air by means of thresholding and morphological refinements. Fixed Hounsfield replacement values were assigned for air (-1000 HU) and soft-tissue (+42 HU), whereas continuous linear mapping was used for bone. RESULTS: The obtained ZTE-derived pseudo-CT images accurately resembled the true CT images (i.e., Dice coefficient for bone overlap of 0.73 ± 0.08 and mean absolute error of 123 ± 25 HU evaluated over the whole head, including errors from residual registration mismatches in the neck and mouth regions). The linear bone mapping accounted for bone density variations. Averaged across five patients, ZTE-based AC demonstrated a PET error of -0.04 ± 1.68% relative to CT-based AC. Similarly, for RTP assessed in eight patients, the absolute dose difference over the target volume was found to be 0.23 ± 0.42%. CONCLUSION: The described method enables MR to pseudo-CT image conversion for the head in an accurate, robust, and fast manner without relying on anatomical prior knowledge. Potential applications include PET/MR-AC, and MR-guided RTP.

Dose painting of prostate cancer based on Gleason score correlations with apparent diffusion coefficients.

BACKGROUND: Gleason scores for prostate cancer correlates with an increased recurrence risk after radiotherapy (RT). Furthermore, higher Gleason scores correlates with decreasing apparent diffusion coefficient (ADC) data from diffusion weighted MRI (DWI-MRI). Based on these observations, we present a formalism for dose painting prescriptions of prostate volumes based on ADC images mapped to Gleason score driven dose-responses. METHODS: The Gleason score driven dose-responses were derived from a learning data set consisting of pre-RT biopsy data and post-RT outcomes for 122 patients treated with a homogeneous dose to the prostate. For a test data set of 18 prostate cancer patients with pre-RT ADC images, we mapped the ADC data to the Gleason driven dose-responses by using probability distributions constructed from published Gleason score correlations with ADC data. We used the Gleason driven dose-responses to optimize dose painting prescriptions that maximize the tumor control probability (TCP) with equal average dose as for the learning sets homogeneous treatment dose. RESULTS: The dose painting prescriptions increased the estimated TCP compared to the homogeneous dose by 0-51% for the learning set and by 4-30% for the test set. The potential for individual TCP gains with dose painting correlated with increasing Gleason score spread and larger prostate volumes. The TCP gains were also found to be larger for patients with a low expected TCP for the homogeneous dose prescription. CONCLUSIONS: We have from retrospective treatment data demonstrated a formalism that yield ADC driven dose painting prescriptions for prostate volumes that potentially can yield significant TCP increases without increasing dose burdens as compared to a homogeneous treatment dose. This motivates further development of the approach to consider more accurate ADC to Gleason mappings, issues with delivery robustness of heterogeneous dose distributions, and patient selection criteria for design of clinical trials.

Perfusion Magnetic Resonance Imaging Changes in Normal Appearing Brain Tissue after Radiotherapy in Glioblastoma Patients may Confound Longitudinal Evaluation of Treatment Response.

BACKGROUND: The aim of this study was assess acute and early delayed radiation-induced changes in normal-appearing brain tissue perfusion as measured with perfusion magnetic resonance imaging (MRI) and the dependence of these changes on the fractionated radiotherapy (FRT) dose level. PATIENTS AND METHODS: Seventeen patients with glioma WHO grade III-IV treated with FRT were included in this prospective study, seven were excluded because of inconsistent FRT protocol or missing examinations. Dynamic susceptibility contrast MRI and contrast-enhanced 3D-T1-weighted (3D-T1w) images were acquired prior to and in average (standard deviation): 3.1 (3.3), 34.4 (9.5) and 103.3 (12.9) days after FRT. Pre-FRT 3D-T1w images were segmented into white- and grey matter. Cerebral blood volume (CBV) and cerebral blood flow (CBF) maps were calculated and co-registered patient-wise to pre-FRT 3D-T1w images. Seven radiation dose regions were created for each tissue type: 0-5 Gy, 5-10 Gy, 10-20 Gy, 20-30 Gy, 30-40 Gy, 40-50 Gy and 50-60 Gy. Mean CBV and CBF were calculated in each dose region and normalised (nCBV and nCBF) to the mean CBV and CBF in 0-5 Gy white- and grey matter reference regions, respectively. RESULTS: Regional and global nCBV and nCBF in white- and grey matter decreased after FRT, followed by a tendency to recover. The response of nCBV and nCBF was dose-dependent in white matter but not in grey matter. CONCLUSIONS: Our data suggest that radiation-induced perfusion changes occur in normal-appearing brain tissue after FRT. This can cause an overestimation of relative tumour perfusion using dynamic susceptibility contrast MRI, and can thus confound tumour treatment evaluation.

Improved quality of computed tomography substitute derived from magnetic resonance (MR) data by incorporation of spatial information--potential application for MR-only radiotherapy and attenuation correction in positron emission tomography.

BACKGROUND: Estimation of computed tomography (CT) equivalent data, i.e. a substitute CT (s-CT), from magnetic resonance (MR) images is a prerequisite both for attenuation correction of positron emission tomography (PET) data acquired with a PET/MR scanner and for dose calculations in an MR-only radiotherapy workflow. It has previously been shown that it is possible to estimate Hounsfield numbers based on MR image intensities, using ultra short echo-time imaging and Gaussian mixture regression (GMR). In the present pilot study we investigate the possibility to also include spatial information in the GMR, with the aim to improve the quality of the s-CT. MATERIAL AND METHODS: MR and CT data for nine patients were used in the present study. For each patient, GMR models were created from the other eight patients, including either both UTE image intensities and spatial information on a voxel by voxel level, or only UTE image intensities. The models were used to create s-CT images for each respective patient. RESULTS: The inclusion of spatial information in the GMR model improved the accuracy of the estimated s-CT. The improvement was most pronounced in smaller, complicated anatomical regions as the inner ear and post-nasal cavities. CONCLUSIONS: This pilot study shows that inclusion of spatial information in GMR models to convert MR data to CT equivalent images is feasible. The accuracy of the s-CT is improved and the spatial information could make it possible to create a general model for the conversion applicable to the whole body.

Evaluation of an attenuation correction method for PET/MR imaging of the head based on substitute CT images.

OBJECT: The aim of this study was to evaluate MR-based attenuation correction of PET emission data of the head, based on a previously described technique that calculates substitute CT (sCT) images from a set of MR images. MATERIALS AND METHODS: Images from eight patients, examined with (18)F-FLT PET/CT and MRI, were included. sCT images were calculated and co-registered to the corresponding CT images, and transferred to the PET/CT scanner for reconstruction. The new reconstructions were then compared with the originals. The effect of replacing bone with soft tissue in the sCT-images was also evaluated. RESULTS: The average relative difference between the sCT-corrected PET images and the CT-corrected PET images was 1.6% for the head and 1.9% for the brain. The average standard deviations of the relative differences within the head were relatively high, at 13.2%, primarily because of large differences in the nasal septa region. For the brain, the average standard deviation was lower, 4.1%. The global average difference in the head when replacing bone with soft tissue was 11%. CONCLUSION: The method presented here has a high rate of accuracy, but high-precision quantitative imaging of the nasal septa region is not possible at the moment.

Towards MR contrast independent synthetic CT generation.

The use of synthetic CT (sCT) in the radiotherapy workflow would reduce costs and scan time while removing the uncertainties around working with both MR and CT modalities. The performance of deep learning (DL) solutions for sCT generation is steadily increasing, however most proposed methods were trained and validated on private datasets of a single contrast from a single scanner. Such solutions might not perform equally well on other datasets, limiting their general usability and therefore value. Additionally, functional evaluations of sCTs such as dosimetric comparisons with CT-based dose calculations better show the impact of the methods, but the evaluations are more labor intensive than pixel-wise metrics. To improve the generalization of an sCT model, we propose to incorporate a pre-trained DL model to pre-process the input MR images by generating artificial proton density, T1 and T2 maps (i.e. contrast-independent quantitative maps), which are then used for sCT generation. Using a dataset of only T2w MR images, the robustness towards input MR contrasts of this approach is compared to a model that was trained using the MR images directly. We evaluate the generated sCTs using pixel-wise metrics and calculating mean radiological depths, as an approximation of the mean delivered dose. On T2w images acquired with the same settings as the training dataset, there was no significant difference between the performance of the models. However, when evaluated on T1w images, and a wide range of other contrasts and scanners from both public and private datasets, our approach outperforms the baseline model. Using a dataset of T2w MR images, our proposed model implements synthetic quantitative maps to generate sCT images, improving the generalization towards other contrasts. Our code and trained models are publicly available.

Hyaluronic acid spacer in prostate cancer radiotherapy: dosimetric effects, spacer stability and long-term toxicity and PRO in a phase II study.

BACKGROUND: Perirectal spacers may be beneficial to reduce rectal side effects from radiotherapy (RT). Here, we present the impact of a hyaluronic acid (HA) perirectal spacer on rectal dose as well as spacer stability, long-term gastrointestinal (GI) and genitourinary (GU) toxicity and patient-reported outcome (PRO). METHODS: In this phase II study 81 patients with low- and intermediate-risk prostate cancer received transrectal injections with HA before external beam RT (78 Gy in 39 fractions). The HA spacer was evaluated with MRI four times; before (MR0) and after HA-injection (MR1), at the middle (MR2) and at the end (MR3) of RT. GI and GU toxicity was assessed by physician for up to five years according to the RTOG scale. PROs were collected using the Swedish National Prostate Cancer Registry and Prostate cancer symptom scale questionnaires. RESULTS: There was a significant reduction in rectal V70% (54.6 Gy) and V90% (70.2 Gy) between MR0 and MR1, as well as between MR0 to MR2 and MR3. From MR1 to MR2/MR3, HA thickness decreased with 28%/32% and CTV-rectum space with 19%/17% in the middle level. The cumulative late grade ≥ 2 GI toxicity at 5 years was 5% and the proportion of PRO moderate or severe overall bowel problems at 5 years follow-up was 12%. Cumulative late grade ≥ 2 GU toxicity at 5 years was 12% and moderate or severe overall urinary problems at 5 years were 10%. CONCLUSION: We show that the HA spacer reduced rectal dose and long-term toxicity.

Region of interest focused MRI to synthetic CT translation using regression and segmentation multi-task network.

Objective. In MR-only clinical workflow, replacing CT with MR image is of advantage for workflow efficiency and reduces radiation to the patient. An important step required to eliminate CT scan from the workflow is to generate the information provided by CT via an MR image. In this work, we aim to demonstrate a method to generate accurate synthetic CT (sCT) from an MR image to suit the radiation therapy (RT) treatment planning workflow. We show the feasibility of the method and make way for a broader clinical evaluation.Approach. We present a machine learning method for sCT generation from zero-echo-time (ZTE) MRI aimed at structural and quantitative accuracies of the image, with a particular focus on the accurate bone density value prediction. The misestimation of bone density in the radiation path could lead to unintended dose delivery to the target volume and results in suboptimal treatment outcome. We propose a loss function that favors a spatially sparse bone region in the image. We harness the ability of the multi-task network to produce correlated outputs as a framework to enable localization of region of interest (RoI) via segmentation, emphasize regression of values within RoI and still retain the overall accuracy via global regression. The network is optimized by a composite loss function that combines a dedicated loss from each task.Main results. We have included 54 brain patient images in this study and tested the sCT images against reference CT on a subset of 20 cases. A pilot dose evaluation was performed on 9 of the 20 test cases to demonstrate the viability of the generated sCT in RT planning. The average quantitative metrics produced by the proposed method over the test set were-(a) mean absolute error (MAE) of 70 ± 8.6 HU; (b) peak signal-to-noise ratio (PSNR) of 29.4 ± 2.8 dB; structural similarity metric (SSIM) of 0.95 ± 0.02; and (d) Dice coefficient of the body region of 0.984 ± 0.Significance. We demonstrate that the proposed method generates sCT images that resemble visual characteristics of a real CT image and has a quantitative accuracy that suits RT dose planning application. We compare the dose calculation from the proposed sCT and the real CT in a radiation therapy treatment planning setup and show that sCT based planning falls within 0.5% target dose error. The method presented here with an initial dose evaluation makes an encouraging precursor to a broader clinical evaluation of sCT based RT planning on different anatomical regions.

The grade of individual prostate cancer lesions predicted by magnetic resonance imaging and positron emission tomography.

BACKGROUND: Multiparametric magnetic resonance imaging (mpMRI) and positron emission tomography (PET) are widely used for the management of prostate cancer (PCa). However, how these modalities complement each other in PCa risk stratification is still largely unknown. We aim to provide insights into the potential of mpMRI and PET for PCa risk stratification. METHODS: We analyzed data from 55 consecutive patients with elevated prostate-specific antigen and biopsy-proven PCa enrolled in a prospective study between December 2016 and December 2019. [68Ga]PSMA-11 PET (PSMA-PET), [11C]Acetate PET (Acetate-PET) and mpMRI were co-registered with whole-mount histopathology. Lower- and higher-grade lesions were defined by International Society of Urological Pathology (ISUP) grade groups (IGG). We used PET and mpMRI data to differentiate between grades in two cases: IGG 3 vs. IGG 2 (case 1) and IGG ≥ 3 vs. IGG ≤ 2 (case 2). The performance was evaluated by receiver operating characteristic (ROC) analysis. RESULTS: We find that the maximum standardized uptake value (SUVmax) for PSMA-PET achieves the highest area under the ROC curve (AUC), with AUCs of 0.72 (case 1) and 0.79 (case 2). Combining the volume transfer constant, apparent diffusion coefficient and T2-weighted images (each normalized to non-malignant prostatic tissue) results in AUCs of 0.70 (case 1) and 0.70 (case 2). Adding PSMA-SUVmax increases the AUCs by 0.09 (p < 0.01) and 0.12 (p < 0.01), respectively. CONCLUSIONS: By co-registering whole-mount histopathology and in-vivo imaging we show that mpMRI and PET can distinguish between lower- and higher-grade prostate cancer, using partially discriminative cut-off values.

Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) are two medical imaging methods commonly used to image prostate cancers. However, the relationship between images obtained with these methods and prostate cancer aggressiveness is not well understood. Here, we investigate whether MRI and PET can differentiate between lower- and higher-grade prostate tumors, where grade is an indicator of how aggressive the disease is likely to be. We find that the characteristics of prostate cancer tumors as seen on MRI and PET scans can help to predict tumor grade. This means that these imaging methods may be helpful when clinicians are predicting patient prognosis and deciding on appropriate treatments. However, further validation is necessary before these approaches are widely implemented for this purpose.

Histopathology-validated lesion detection rates of clinically significant prostate cancer with mpMRI, [ 68 Ga]PSMA-11-PET and [ 11 C]Acetate-PET.

OBJECTIVE: PET/CT and multiparametric MRI (mpMRI) are important diagnostic tools in clinically significant prostate cancer (csPC). The aim of this study was to compare csPC detection rates with [ 68 Ga]PSMA-11-PET (PSMA)-PET, [ 11 C]Acetate (ACE)-PET, and mpMRI with histopathology as reference, to identify the most suitable imaging modalities for subsequent hybrid imaging. An additional aim was to compare inter-reader variability to assess reproducibility. METHODS: During 2016-2019, all study participants were examined with PSMA-PET/mpMRI and ACE-PET/CT prior to radical prostatectomy. PSMA-PET, ACE-PET and mpMRI were evaluated separately by two observers, and were compared with histopathology-defined csPC. Statistical analyses included two-sided McNemar test and index of specific agreement. RESULTS: Fifty-five study participants were included, with 130 histopathological intraprostatic lesions >0.05 cc. Of these, 32% (42/130) were classified as csPC with ISUP grade ≥2 and volume >0.5 cc. PSMA-PET and mpMRI showed no difference in performance ( P  = 0.48), with mean csPC detection rate of 70% (29.5/42) and 74% (31/42), respectively, while with ACE-PET the mean csPC detection rate was 37% (15.5/42). Interobserver agreement was higher with PSMA-PET compared to mpMRI [79% (26/33) vs 67% (24/38)]. Including all detected lesions from each pair of observers, the detection rate increased to 90% (38/42) with mpMRI, and 79% (33/42) with PSMA-PET. CONCLUSION: PSMA-PET and mpMRI showed high csPC detection rates and superior performance compared to ACE-PET. The interobserver agreement indicates higher reproducibility with PSMA-PET. The combined result of all observers in both PSMA-PET and mpMRI showed the highest detection rate, suggesting an added value of a hybrid imaging approach.

Voxel-wise uncertainty in CT substitute derived from MRI.

PURPOSE: In an earlier work, we demonstrated that substitutes for CT images can be derived from MR images using ultrashort echo time (UTE) sequences, conventional T2 weighted sequences, and Gaussian mixture regression (GMR). In this study, we extend this work by analyzing the uncertainties associated with the GMR model and the information contributions from the individual imaging sequences. METHODS: An analytical expression for the voxel-wise conditional expected absolute deviation (EAD) in substitute CT (s-CT) images was derived. The expression depends only on MR images and can thus be calculated along with each s-CT image. The uncertainty measure was evaluated by comparing the EAD to the true mean absolute prediction deviation (MAPD) between the s-CT and CT images for 14 patients. Further, the influence of the different MR images included in the GMR model on the generated s-CTs was investigated by removing one or more images and evaluating the MAPD for a spectrum of predicted radiological densities. RESULTS: The largest EAD was predicted at air-soft tissue and bone-soft tissue interfaces. The EAD agreed with the MAPD in both these regions and in regions with lower EADs, such as the brain. Two of the MR images included in the GMR model were found to be mutually redundant for the purpose of s-CT generation. CONCLUSIONS: The presented uncertainty estimation method accurately predicts the voxel-wise MAPD in s-CT images. Also, the non-UTE sequence previously used in the model was found to be redundant.

A Data-Adaptive Loss Function for Incomplete Data and Incremental Learning in Semantic Image Segmentation.

In the last years, deep learning has dramatically improved the performances in a variety of medical image analysis applications. Among different types of deep learning models, convolutional neural networks have been among the most successful and they have been used in many applications in medical imaging. Training deep convolutional neural networks often requires large amounts of image data to generalize well to new unseen images. It is often time-consuming and expensive to collect large amounts of data in the medical image domain due to expensive imaging systems, and the need for experts to manually make ground truth annotations. A potential problem arises if new structures are added when a decision support system is already deployed and in use. Since the field of radiation therapy is constantly developing, the new structures would also have to be covered by the decision support system. In the present work, we propose a novel loss function to solve multiple problems: imbalanced datasets, partially-labeled data, and incremental learning. The proposed loss function adapts to the available data in order to utilize all available data, even when some have missing annotations. We demonstrate that the proposed loss function also works well in an incremental learning setting, where an existing model is easily adapted to semi-automatically incorporate delineations of new organs when they appear. Experiments on a large in-house dataset show that the proposed method performs on par with baseline models, while greatly reducing the training time and eliminating the hassle of maintaining multiple models in practice.

QU-BraTS: MICCAI BraTS 2020 Challenge on Quantifying Uncertainty in Brain Tumor Segmentation - Analysis of Ranking Scores and Benchmarking Results.

Deep learning (DL) models have provided state-of-the-art performance in various medical imaging benchmarking challenges, including the Brain Tumor Segmentation (BraTS) challenges. However, the task of focal pathology multi-compartment segmentation (e.g., tumor and lesion sub-regions) is particularly challenging, and potential errors hinder translating DL models into clinical workflows. Quantifying the reliability of DL model predictions in the form of uncertainties could enable clinical review of the most uncertain regions, thereby building trust and paving the way toward clinical translation. Several uncertainty estimation methods have recently been introduced for DL medical image segmentation tasks. Developing scores to evaluate and compare the performance of uncertainty measures will assist the end-user in making more informed decisions. In this study, we explore and evaluate a score developed during the BraTS 2019 and BraTS 2020 task on uncertainty quantification (QU-BraTS) and designed to assess and rank uncertainty estimates for brain tumor multi-compartment segmentation. This score (1) rewards uncertainty estimates that produce high confidence in correct assertions and those that assign low confidence levels at incorrect assertions, and (2) penalizes uncertainty measures that lead to a higher percentage of under-confident correct assertions. We further benchmark the segmentation uncertainties generated by 14 independent participating teams of QU-BraTS 2020, all of which also participated in the main BraTS segmentation task. Overall, our findings confirm the importance and complementary value that uncertainty estimates provide to segmentation algorithms, highlighting the need for uncertainty quantification in medical image analyses. Finally, in favor of transparency and reproducibility, our evaluation code is made publicly available at https://github.com/RagMeh11/QU-BraTS.

CT substitute derived from MRI sequences with ultrashort echo time.

PURPOSE: Methods for deriving computed tomography (CT) equivalent information from MRI are needed for attenuation correction in PET/MRI applications, as well as for patient positioning and dose planning in MRI based radiation therapy workflows. This study presents a method for generating a drop in substitute for a CT image from a set of magnetic resonance (MR)images. METHODS: A Gaussian mixture regression model was used to link the voxel values in CT images to the voxel values in images from three MRI sequences: one T2 weighted 3D spin echo based sequence and two dual echo ultrashort echo time MRI sequences with different echo times and flip angles. The method used a training set of matched MR and CT data that after training was able to predict a substitute CT (s-CT) based entirely on the MR information for a new patient. Method validation was achieved using datasets covering the heads of five patients and applying leave-one-out cross-validation (LOOCV). During LOOCV, the model was estimated from the MR and CT data of four patients (training set) and applied to the MR data of the remaining patient (validation set) to generate an s-CT image. This procedure was repeated for all five training and validation data combinations. RESULTS: The mean absolute error for the CT number in the s-CT images was 137 HU. No large differences in method accuracy were noted for the different patients, indicating a robust method. The largest errors in the s-CT images were found at air-tissue and bone-tissue interfaces. The model accurately discriminated between air and bone, as well as between soft tissues and nonsoft tissues. CONCLUSIONS: The s-CT method has the potential to provide an accurate estimation of CT information without risk of geometrical inaccuracies as the model is voxel based. Therefore, s-CT images could be well suited as alternatives to CT images for dose planning in radiotherapy and attenuation correction in PET/MRI.

Intrafractional motion models based on principal components in Magnetic Resonance guided prostate radiotherapy.

BACKGROUND AND PURPOSE: Devices that combine an MR-scanner with a Linac for radiotherapy, referred to as MR-Linac systems, introduce the possibility to acquire high resolution images prior and during treatment. Hence, there is a possibility to acquire individualised learning sets for motion models for each fraction and the construction of intrafractional motion models. We investigated the feasibility for a principal component analysis (PCA) based, intrafractional motion model of the male pelvic region. MATERIALS AND METHODS: 4D-scans of nine healthy male volunteers were utilized, FOV covering the entire pelvic region including prostate, bladder and rectum with manual segmentation of each organ at each time frame. Deformable image registration with an optical flow algorithm was performed for each subject with the first time frame as reference. PCA was performed on a subset of the resulting displacement vector fields to construct individualised motion models evaluated on the remaining fields. RESULTS: The registration algorithm produced accurate registration result, in general DICE overlap > 0.95 across all time frames. Cumulative variance of the eigen values from the PCA showed that 50% or more of the motion is explained in the first component for all subjects. However, the size and direction for the components differed between subjects. Adding more than two components did not improve the accuracy significantly and the model was able to explain motion down to about 1 mm. CONCLUSIONS: An individualised intrafractional male pelvic motion model is feasible. Geometric accuracy was about 1 mm based on 1-2 principal components.

Artificial intelligence and the medical physics profession - A Swedish perspective.

BACKGROUND: There is a continuous and dynamic discussion on artificial intelligence (AI) in present-day society. AI is expected to impact on healthcare processes and could contribute to a more sustainable use of resources allocated to healthcare in the future. The aim for this work was to establish a foundation for a Swedish perspective on the potential effect of AI on the medical physics profession. MATERIALS AND METHODS: We designed a survey to gauge viewpoints regarding AI in the Swedish medical physics community. Based on the survey results and present-day situation in Sweden, a SWOT analysis was performed on the implications of AI for the medical physics profession. RESULTS: Out of 411 survey recipients, 163 responded (40%). The Swedish medical physicists with a professional license believed (90%) that AI would change the practice of medical physics but did not foresee (81%) that AI would pose a risk to their practice and career. The respondents were largely positive to the inclusion of AI in educational programmes. According to self-assessment, the respondents' knowledge of and workplace preparedness for AI was generally low. CONCLUSIONS: From the survey and SWOT analysis we conclude that AI will change the medical physics profession and that there are opportunities for the profession associated with the adoption of AI in healthcare. To overcome the weakness of limited AI knowledge, potentially threatening the role of medical physicists, and build upon the strong position in Swedish healthcare, medical physics education and training should include learning objectives on AI.