Cross-modal tumor segmentation using generative blending augmentation and self-training.

OBJECTIVES: Data scarcity and domain shifts lead to biased training sets that do not accurately represent deployment conditions. A related practical problem is cross-modal image segmentation, where the objective is to segment unlabelled images using previously labelled datasets from other imaging modalities. METHODS: We propose a cross-modal segmentation method based on conventional image synthesis boosted by a new data augmentation technique called Generative Blending Augmentation (GBA). GBA leverages a SinGAN model to learn representative generative features from a single training image to diversify realistically tumor appearances. This way, we compensate for image synthesis errors, subsequently improving the generalization power of a downstream segmentation model. The proposed augmentation is further combined to an iterative self-training procedure leveraging pseudo labels at each pass. RESULTS: The proposed solution ranked first for vestibular schwannoma (VS) segmentation during the validation and test phases of the MICCAI CrossMoDA 2022 challenge, with best mean Dice similarity and average symmetric surface distance measures. CONCLUSION AND SIGNIFICANCE: Local contrast alteration of tumor appearances and iterative self-training with pseudo labels are likely to lead to performance improvements in a variety of segmentation contexts.

Contrastive image adaptation for acquisition shift reduction in medical imaging.

The domain shift, or acquisition shift in medical imaging, is responsible for potentially harmful differences between development and deployment conditions of medical image analysis techniques. There is a growing need in the community for advanced methods that could mitigate this issue better than conventional approaches. In this paper, we consider configurations in which we can expose a learning-based pixel level adaptor to a large variability of unlabeled images during its training, i.e. sufficient to span the acquisition shift expected during the training or testing of a downstream task model. We leverage the ability of convolutional architectures to efficiently learn domain-agnostic features and train a many-to-one unsupervised mapping between a source collection of heterogeneous images from multiple unknown domains subjected to the acquisition shift and a homogeneous subset of this source set of lower cardinality, potentially constituted of a single image. To this end, we propose a new cycle-free image-to-image architecture based on a combination of three loss functions : a contrastive PatchNCE loss, an adversarial loss and an edge preserving loss allowing for rich domain adaptation to the target image even under strong domain imbalance and low data regimes. Experiments support the interest of the proposed contrastive image adaptation approach for the regularization of downstream deep supervised segmentation and cross-modality synthesis models.

PET/CT-Based Radiogenomics Supports KEAP1/NFE2L2 Pathway Targeting for Non-Small Cell Lung Cancer Treated with Curative Radiotherapy.

In lung cancer patients, radiotherapy is associated with a increased risk of local relapse (LR) when compared with surgery but with a preferable toxicity profile. The KEAP1/NFE2L2 mutational status (MutKEAP1/NFE2L2) is significantly correlated with LR in patients treated with radiotherapy but is rarely available. Prediction of MutKEAP1/NFE2L2 with noninvasive modalities could help to further personalize each therapeutic strategy. Methods: Based on a public cohort of 770 patients, model RNA (M-RNA) was first developed using continuous gene expression levels to predict MutKEAP1/NFE2L2, resulting in a binary output. The model PET/CT (M-PET/CT) was then built to predict M-RNA binary output using PET/CT-extracted radiomics features. M-PET/CT was validated on an external cohort of 151 patients treated with curative volumetric modulated arc radiotherapy. Each model was built, internally validated, and evaluated on a separate cohort using a multilayer perceptron network approach. Results: The M-RNA resulted in a C statistic of 0.82 in the testing cohort. With a training cohort of 101 patients, the retained M-PET/CT resulted in an area under the curve of 0.90 (P < 0.001). With a probability threshold of 20% applied to the testing cohort, M-PET/CT achieved a C statistic of 0.7. The same radiomics model was validated on the volumetric modulated arc radiotherapy cohort as patients were significantly stratified on the basis of their risk of LR with a hazard ratio of 2.61 (P = 0.02). Conclusion: Our approach enables the prediction of MutKEAP1/NFE2L2 using PET/CT-extracted radiomics features and efficiently classifies patients at risk of LR in an external cohort treated with radiotherapy.

Multi-modal medical Transformers: A meta-analysis for medical image segmentation in oncology.

Multi-modal medical image segmentation is a crucial task in oncology that enables the precise localization and quantification of tumors. The aim of this work is to present a meta-analysis of the use of multi-modal medical Transformers for medical image segmentation in oncology, specifically focusing on multi-parametric MR brain tumor segmentation (BraTS2021), and head and neck tumor segmentation using PET-CT images (HECKTOR2021). The multi-modal medical Transformer architectures presented in this work exploit the idea of modality interaction schemes based on visio-linguistic representations: (i) single-stream, where modalities are jointly processed by one Transformer encoder, and (ii) multiple-stream, where the inputs are encoded separately before being jointly modeled. A total of fourteen multi-modal architectures are evaluated using different ranking strategies based on dice similarity coefficient (DSC) and average symmetric surface distance (ASSD) metrics. In addition, cost indicators such as the number of trainable parameters and the number of multiply-accumulate operations (MACs) are reported. The results demonstrate that multi-path hybrid CNN-Transformer-based models improve segmentation accuracy when compared to traditional methods, but come at the cost of increased computation time and potentially larger model size.

Learn2Reg: Comprehensive Multi-Task Medical Image Registration Challenge, Dataset and Evaluation in the Era of Deep Learning.

Image registration is a fundamental medical image analysis task, and a wide variety of approaches have been proposed. However, only a few studies have comprehensively compared medical image registration approaches on a wide range of clinically relevant tasks. This limits the development of registration methods, the adoption of research advances into practice, and a fair benchmark across competing approaches. The Learn2Reg challenge addresses these limitations by providing a multi-task medical image registration data set for comprehensive characterisation of deformable registration algorithms. A continuous evaluation will be possible at https://learn2reg.grand-challenge.org. Learn2Reg covers a wide range of anatomies (brain, abdomen, and thorax), modalities (ultrasound, CT, MR), availability of annotations, as well as intra- and inter-patient registration evaluation. We established an easily accessible framework for training and validation of 3D registration methods, which enabled the compilation of results of over 65 individual method submissions from more than 20 unique teams. We used a complementary set of metrics, including robustness, accuracy, plausibility, and runtime, enabling unique insight into the current state-of-the-art of medical image registration. This paper describes datasets, tasks, evaluation methods and results of the challenge, as well as results of further analysis of transferability to new datasets, the importance of label supervision, and resulting bias. While no single approach worked best across all tasks, many methodological aspects could be identified that push the performance of medical image registration to new state-of-the-art performance. Furthermore, we demystified the common belief that conventional registration methods have to be much slower than deep-learning-based methods.

Combination of Radiomics Features and Functional Radiosensitivity Enhances Prediction of Acute Pulmonary Toxicity in a Prospective Validation Cohort of Patients with a Locally Advanced Lung Cancer Treated with VMAT-Radiotherapy.

INTRODUCTION: The standard of care for people with locally advanced lung cancer (LALC) who cannot be operated on is (chemo)-radiation. Despite the application of dose constraints, acute pulmonary toxicity (APT) still often occurs. Prediction of APT is of paramount importance for the development of innovative therapeutic combinations. The two models were previously individually created. With success, the Rad-model incorporated six radiomics functions. After additional validation in prospective cohorts, a Pmap-model was created by identifying a specific region of the right posterior lung and incorporating several clinical and dosimetric parameters. To create and test a novel model to forecast the risk of APT in two cohorts receiving volumetric arctherapy radiotherapy (VMAT), we aimed to include all the variables in this study. METHODS: In the training cohort, we retrospectively included all patients treated by VMAT for LALC at one institution between 2015 and 2018. APT was assessed according to the CTCAE v4.0 scale. Usual clinical and dosimetric features, as well as the mean dose to the pre-defined Pmap zone (DMeanPmap), were processed using a neural network approach and subsequently validated on an observational prospective cohort. The model was evaluated using the area under the curve (AUC) and balanced accuracy (Bacc). RESULTS: 165 and 42 patients were enrolled in the training and test cohorts, with APT rates of 22.4 and 19.1%, respectively. The AUCs for the Rad and Pmap models in the validation cohort were 0.83 and 0.81, respectively, whereas the AUC for the combined model (Comb-model) was 0.90. The Bacc for the Rad, Pmap, and Comb models in the validation cohort were respectively 78.7, 82.4, and 89.7%. CONCLUSION: The accuracy of prediction models were increased by combining radiomics, DMeanPmap, and common clinical and dosimetric features. The use of this model may improve the evaluation of APT risk and provide access to novel therapeutic alternatives, such as dose escalation or creative therapy combinations.

VMAT-Based Planning Allows Sparing of a Spatial Dose Pattern Associated with Radiation Pneumonitis in Patients Treated with Radiotherapy for a Locally Advanced Lung Cancer.

Introduction: In patients treated with radiotherapy for locally advanced lung cancer, respect for dose constraints to organs at risk (OAR) insufficiently protects patients from acute pulmonary toxicity (APT), such toxicities being associated with a potential impact on the treatment's completion and the patient's quality of life. Dosimetric planning does not take into account regional lung functionality. An APT prediction model combining usual dosimetry features with the mean dose (DMeanPmap) received by a voxel-based volume (Pmap) localized in the posterior right lung has been previously developed. A DMeanPmap of ≥30.3 Gy or a predicted APT probability (ProbAPT) of ≥8% were associated with a higher risk of APT. In the present study, the authors aim to demonstrate the possibility of decreasing the DMeanPmap via a volumetric arctherapy (VMAT)-based adapted planning and evaluate the impact on the risk of APT. Methods: Among the 207 patients included in the initial study, only patients who presented with APT of ≥grade 2 and with a probability of APT ≥ 8% based on the prediction model were included. Dosimetry planning was optimized with a new constraint (DMeanPmap < 30.3 Gy) added to the usual constraints. The initial and optimized treatment plans were compared using the t-test for the independent variables and the non-parametric Mann−Whitney U test otherwise, regarding both doses to the OARs and PTV (Planning Target Volume) coverage. Conformity and heterogeneity indexes were also compared. The risk of APT was recalculated using the new dosimetric features and the APT prediction model. Results: Dosimetric optimization was considered successful for 27 out of the 44 included patients (61.4%), meaning the dosimetric constraint on the Pmap region was achieved without compromising the PTV coverage (p = 0.61). The optimization significantly decreased the median DMeanPmap from 28.8 Gy (CI95% 24.2−33.4) to 22.1 Gy (CI95% 18.3−26.0). When recomputing the risk of APT using the new dosimetric features, the optimization significantly reduced the risk of APT (p < 0.0001) by reclassifying 43.2% (19/44) of the patients. Conclusion: Our approach appears to be both easily implementable on a daily basis and efficient at reducing the risk of APT. Regional radiosensitivity should be considered in usual lung dose constraints, opening the possibility of new treatment strategies, such as dose escalation or innovative treatment associations.

Deep learning-based segmentation in prostate radiation therapy using Monte Carlo simulated cone-beam computed tomography.

PURPOSE: Segmenting organs in cone-beam CT (CBCT) images would allow to adapt the radiotherapy based on the organ deformations that may occur between treatment fractions. However, this is a difficult task because of the relative lack of contrast in CBCT images, leading to high inter-observer variability. Deformable image registration (DIR) and deep-learning based automatic segmentation approaches have shown interesting results for this task in the past years. However, they are either sensitive to large organ deformations, or require to train a convolutional neural network (CNN) from a database of delineated CBCT images, which is difficult to do without improvement of image quality. In this work, we propose an alternative approach: to train a CNN (using a deep learning-based segmentation tool called nnU-Net) from a database of artificial CBCT images simulated from planning CT, for which it is easier to obtain the organ contours. METHODS: Pseudo-CBCT (pCBCT) images were simulated from readily available segmented planning CT images, using the GATE Monte Carlo simulation. CT reference delineations were copied onto the pCBCT, resulting in a database of segmented images used to train the neural network. The studied segmentation contours were: bladder, rectum, and prostate contours. We trained multiple nnU-Net models using different training: (1) segmented real CBCT, (2) pCBCT, (3) segmented real CT and tested on pseudo-CT (pCT) generated from CBCT with cycleGAN, and (4) a combination of (2) and (3). The evaluation was performed on different datasets of segmented CBCT or pCT by comparing predicted segmentations with reference ones thanks to Dice similarity score and Hausdorff distance. A qualitative evaluation was also performed to compare DIR-based and nnU-Net-based segmentations. RESULTS: Training with pCBCT was found to lead to comparable results to using real CBCT images. When evaluated on CBCT obtained from the same hospital as the CT images used in the simulation of the pCBCT, the model trained with pCBCT scored mean DSCs of 0.92 ± 0.05, 0.87 ± 0.02, and 0.85 ± 0.04 and mean Hausdorff distance 4.67 ± 3.01, 3.91 ± 0.98, and 5.00 ± 1.32 for the bladder, rectum, and prostate contours respectively, while the model trained with real CBCT scored mean DSCs of 0.91 ± 0.06, 0.83 ± 0.07, and 0.81 ± 0.05 and mean Hausdorff distance 5.62 ± 3.24, 6.43 ± 5.11, and 6.19 ± 1.14 for the bladder, rectum, and prostate contours, respectively. It was also found to outperform models using pCT or a combination of both, except for the prostate contour when tested on a dataset from a different hospital. Moreover, the resulting segmentations demonstrated a clinical acceptability, where 78% of bladder segmentations, 98% of rectum segmentations, and 93% of prostate segmentations required minor or no corrections, and for 76% of the patients, all structures of the patient required minor or no corrections. CONCLUSION: We proposed to use simulated CBCT images to train a nnU-Net segmentation model, avoiding the need to gather complex and time-consuming reference delineations on CBCT images.

Prediction of recurrence after surgery in colorectal cancer patients using radiomics from diagnostic contrast-enhanced computed tomography: a two-center study.

OBJECTIVES: To assess the value of contrast-enhanced (CE) diagnostic CT scans characterized through radiomics as predictors of recurrence for patients with stage II and III colorectal cancer in a two-center context. MATERIALS AND METHODS: This study included 193 patients diagnosed with stage II and III colorectal adenocarcinoma from 1 July 2008 to 15 March 2017 in two different French University Hospitals. To compensate for the variability in two-center data, a statistical harmonization method Bootstrapped ComBat (B-ComBat) was used. Models predicting disease-free survival (DFS) were built using 3 different machine learning (ML): (1) multivariate regression (MR) with 10-fold cross-validation after feature selection based on least absolute shrinkage and selection operator (LASSO), (2) random forest (RF), and (3) support vector machine (SVM), both with embedded feature selection. RESULTS: The performance for both balanced and 95% sensitivity models was systematically higher after our proposed B-ComBat harmonization compared to the use of the original untransformed data. The most clinically relevant performance was achieved by the multivariate regression model combining a clinical variable (postoperative chemotherapy) with two radiomics shape descriptors (compactness and least axis length) with a BAcc of 0.78 and an MCC of 0.6 associated with a required sensitivity of 95%. The resulting stratification in terms of DFS was significant (p = 0.00021), especially compared to the use of unharmonized original data (p = 0.17). CONCLUSIONS: Radiomics models derived from contrast-enhanced CT could be trained and validated in a two-center cohort with a good predictive performance of recurrence in stage II et III colorectal cancer patients. KEY POINTS: • Adjuvant therapy decision in colorectal cancer can be a challenge in medical oncology. • Radiomics models, derived from diagnostic CT, trained and validated in a two-center cohort, could predict recurrence in stage II and III colorectal cancer patients. • Identifying patients with a low risk of recurrence, these models could facilitate treatment optimization and avoid unnecessary treatment.

Development of a Radiomic-Based Model Predicting Lymph Node Involvement in Prostate Cancer Patients.

Significant advances in lymph node involvement (LNI) risk modeling in prostate cancer (PCa) have been achieved with the addition of visual interpretation of magnetic resonance imaging (MRI) data, but it is likely that quantitative analysis could further improve prediction models. In this study, we aimed to develop and internally validate a novel LNI risk prediction model based on radiomic features extracted from preoperative multimodal MRI. All patients who underwent a preoperative MRI and radical prostatectomy with extensive lymph node dissection were retrospectively included in a single institution. Patients were randomly divided into the training (60%) and testing (40%) sets. Radiomic features were extracted from the index tumor volumes, delineated on the apparent diffusion coefficient corrected map and the T2 sequences. A ComBat harmonization method was applied to account for inter-site heterogeneity. A prediction model was trained using a neural network approach (Multilayer Perceptron Network, SPSS v24.0©) combining clinical, radiomic and all features. It was then evaluated on the testing set and compared to the current available models using the Receiver Operative Characteristics and the C-Index. Two hundred and eighty patients were included, with a median age of 65.2 y (45.3-79.6), a mean PSA level of 9.5 ng/mL (1.04-63.0) and 79.6% of ISUP ≥ 2 tumors. LNI occurred in 51 patients (18.2%), with a median number of extracted nodes of 15 (10-19). In the testing set, with their respective cutoffs applied, the Partin, Roach, Yale, MSKCC, Briganti 2012 and 2017 models resulted in a C-Index of 0.71, 0.66, 0.55, 0.67, 0.65 and 0.73, respectively, while our proposed combined model resulted in a C-Index of 0.89 in the testing set. Radiomic features extracted from the preoperative MRI scans and combined with clinical features through a neural network seem to provide added predictive performance compared to state of the art models regarding LNI risk prediction in PCa.

Development and prospective validation of a spatial dose pattern based model predicting acute pulmonary toxicity in patients treated with volumetric arc-therapy for locally advanced lung cancer.

INTRODUCTION: (Chemo)-radiotherapy is the standard treatment for patients with locally advanced lung cancer (LALC) not accessible to surgery. Despite strict application of dose constraints, acute toxicities such as acute pulmonary toxicity (APT) remain frequent, and may impact treatment's compliance and patients' quality of life. Previously, on a population treated with intensity-modulated photon therapy or passive scattering proton therapy, spatial dose patterns associated with APT were identified in the lower lungs, especially in the posterior right lung. In the present study, we aim to define these spatial dose patterns on a retrospective cohort treated by volumetric-arctherapy (VMAT) and to validate our findings prospectively. METHODS: For the training cohort, we retrospectively included all patients treated in our institution by VMAT for a LALC between 2015 and 2018. APT was scored according to the CTCAE v4.0 scale. All dose maps were registered to a thorax phantom using a segmentation-based elastic registration. Voxel-based analysis of local dose differences was performed with a non-parametric permutation test accounting for n = 10.000 permutations, producing a 3-dimensional significance maps on which clusters of voxels that exhibited significant dose differences (p < 0.05) between the two toxicity groups (APT ≥ grade 2 vs APT < grade 2) were identified. A prediction model (Pmap-Model) was then built using a neural network approach and then applied to an observational prospective cohort for validation. The model was evaluated using the Area under the curve (AUC) and the balanced accuracy (Bacc: mean of the sensitivity and specificity). RESULTS: 165 and 42 patients were included in the training and validation cohorts, with respective APT rates of 22.4% and 19.1%. In the training cohort, a cluster of voxels (Pmap-region) was identified in the posterior right lung. In the training cohort, the Pmap-Model combining 11 features among which the mean dose to the Pmap-region resulted in an AUC of 0.99 and a Bacc of 99.2 using an 8% probability threshold. Using the same voxel cluster on the validation cohort, the Pmap-model resulted in an AUC of 0.81 and a Bacc of 82.0. CONCLUSION: Our APT-prediction model was successfully validated in a prospective cohort treated by VMAT. Regional radiosensitivity should be considered in usual lung dose constraints, opening the possibility of easily implementable adaptive dosimetry planning.

Radiomics Analysis of 3D Dose Distributions to Predict Toxicity of Radiotherapy for Cervical Cancer.

Standard treatment for locally advanced cervical cancer (LACC) is chemoradiotherapy followed by brachytherapy. Despite radiation therapy advances, the toxicity rate remains significant. In this study, we compared the prediction of toxicity events after radiotherapy for locally advanced cervical cancer (LACC), based on either dose-volume histogram (DVH) parameters or the use of a radiomics approach applied to dose maps at the voxel level. Toxicity scores using the Common Terminology Criteria for Adverse Events (CTCAE v4), spatial dose distributions, and usual clinical predictors for the toxicity of 102 patients treated with chemoradiotherapy followed by brachytherapy for LACC were used in this study. In addition to usual DVH parameters, 91 radiomic features were extracted from rectum, bladder and vaginal 3D dose distributions, after discretization into a fixed bin width of 1 Gy. They were evaluated for predictive modelling of rectal, genitourinary (GU) and vaginal toxicities (grade ≥ 2). Logistic Normal Tissue Complication Probability (NTCP) models were derived using clinical parameters only or combinations of clinical, DVH and radiomics. For rectal acute/late toxicities, the area under the curve (AUC) using clinical parameters was 0.53/0.65, which increased to 0.66/0.63, and 0.76/0.87, with the addition of DVH or radiomics parameters, respectively. For GU acute/late toxicities, the AUC increased from 0.55/0.56 (clinical only) to 0.84/0.90 (+DVH) and 0.83/0.96 (clinical + DVH + radiomics). For vaginal acute/late toxicities, the AUC increased from 0.51/0.57 (clinical only) to 0.58/0.72 (+DVH) and 0.82/0.89 (clinical + DVH + radiomics). The predictive performance of NTCP models based on radiomics features was higher than the commonly used clinical and DVH parameters. Dosimetric radiomics analysis is a promising tool for NTCP modelling in radiotherapy.

Dosimetric Validation of a GAN-Based Pseudo-CT Generation for MRI-Only Stereotactic Brain Radiotherapy.

PURPOSE: Stereotactic radiotherapy (SRT) has become widely accepted as a treatment of choice for patients with a small number of brain metastases that are of an acceptable size, allowing for better target dose conformity, resulting in high local control rates and better sparing of organs at risk. An MRI-only workflow could reduce the risk of misalignment between magnetic resonance imaging (MRI) brain studies and computed tomography (CT) scanning for SRT planning, while shortening delays in planning. Given the absence of a calibrated electronic density in MRI, we aimed to assess the equivalence of synthetic CTs generated by a generative adversarial network (GAN) for planning in the brain SRT setting. METHODS: All patients with available MRIs and treated with intra-cranial SRT for brain metastases from 2014 to 2018 in our institution were included. After co-registration between the diagnostic MRI and the planning CT, a synthetic CT was generated using a 2D-GAN (2D U-Net). Using the initial treatment plan (Pinnacle v9.10, Philips Healthcare), dosimetric comparison was performed using main dose-volume histogram (DVH) endpoints in respect to ICRU 91 guidelines (Dmax, Dmean, D2%, D50%, D98%) as well as local and global gamma analysis with 1%/1 mm, 2%/1 mm and 2%/2 mm criteria and a 10% threshold to the maximum dose. t-test analysis was used for comparison between the two cohorts (initial and synthetic dose maps). RESULTS: 184 patients were included, with 290 treated brain metastases. The mean number of treated lesions per patient was 1 (range 1-6) and the median planning target volume (PTV) was 6.44 cc (range 0.12-45.41). Local and global gamma passing rates (2%/2 mm) were 99.1 CI95% (98.1-99.4) and 99.7 CI95% (99.6-99.7) respectively (CI: confidence interval). DVHs were comparable, with no significant statistical differences regarding ICRU 91's endpoints. CONCLUSIONS: Our study is the first to compare GAN-generated CT scans from diagnostic brain MRIs with initial CT scans for the planning of brain stereotactic radiotherapy. We found high similarity between the planning CT and the synthetic CT for both the organs at risk and the target volumes. Prospective validation is under investigation at our institution.

Convolutional neural networks for PET functional volume fully automatic segmentation: development and validation in a multi-center setting.

PURPOSE: In this work, we addressed fully automatic determination of tumor functional uptake from positron emission tomography (PET) images without relying on other image modalities or additional prior constraints, in the context of multicenter images with heterogeneous characteristics. METHODS: In cervical cancer, an additional challenge is the location of the tumor uptake near or even stuck to the bladder. PET datasets of 232 patients from five institutions were exploited. To avoid unreliable manual delineations, the ground truth was generated with a semi-automated approach: a volume containing the tumor and excluding the bladder was first manually determined, then a well-validated, semi-automated approach relying on the Fuzzy locally Adaptive Bayesian (FLAB) algorithm was applied to generate the ground truth. Our model built on the U-Net architecture incorporates residual blocks with concurrent spatial squeeze and excitation modules, as well as learnable non-linear downsampling and upsampling blocks. Experiments relied on cross-validation (four institutions for training and validation, and the fifth for testing). RESULTS: The model achieved good Dice similarity coefficient (DSC) with little variability across institutions (0.80 ± 0.03), with higher recall (0.90 ± 0.05) than precision (0.75 ± 0.05) and improved results over the standard U-Net (DSC 0.77 ± 0.05, recall 0.87 ± 0.02, precision 0.74 ± 0.08). Both vastly outperformed a fixed threshold at 40% of SUVmax (DSC 0.33 ± 0.15, recall 0.52 ± 0.17, precision 0.30 ± 0.16). In all cases, the model could determine the tumor uptake without including the bladder. Neither shape priors nor anatomical information was required to achieve efficient training. CONCLUSION: The proposed method could facilitate the deployment of a fully automated radiomics pipeline in such a challenging multicenter context.

Radiomics in PET/CT: Current Status and Future AI-Based Evolutions.

This short review aims at providing the readers with an update on the current status, as well as future perspectives in the quickly evolving field of radiomics applied to the field of PET/CT imaging. Numerous pitfalls have been identified in study design, data acquisition, segmentation, features calculation and modeling by the radiomics community, and these are often the same issues across all image modalities and clinical applications, however some of these are specific to PET/CT (and SPECT/CT) imaging and therefore the present paper focuses on those. In most cases, recommendations and potential methodological solutions do exist and should therefore be followed to improve the overall quality and reproducibility of published studies. In terms of future evolutions, the techniques from the larger field of artificial intelligence (AI), including those relying on deep neural networks (also known as deep learning) have already shown impressive potential to provide solutions, especially in terms of automation, but also to maybe fully replace the tools the radiomics community has been using until now in order to build the usual radiomics workflow. Some important challenges remain to be addressed before the full impact of AI may be realized but overall the field has made striking advances over the last few years and it is expected advances will continue at a rapid pace.

Artificial intelligence, machine (deep) learning and radio(geno)mics: definitions and nuclear medicine imaging applications.

Techniques from the field of artificial intelligence, and more specifically machine (deep) learning methods, have been core components of most recent developments in the field of medical imaging. They are already being exploited or are being considered to tackle most tasks, including image reconstruction, processing (denoising, segmentation), analysis and predictive modelling. In this review we introduce and define these key concepts and discuss how the techniques from this field can be applied to nuclear medicine imaging applications with a particular focus on radio(geno)mics.

Prostate Volume Segmentation in TRUS Using Hybrid Edge-Bhattacharyya Active Surfaces.

OBJECTIVE: We present a new hybrid edge and region-based parametric deformable model, or active surface, for prostate volume segmentation in transrectal ultrasound (TRUS) images. METHODS: Our contribution is threefold. First, we develop a new edge detector derived from the radial bas-relief approach, allowing for better scalar prostate edge detection in low contrast configurations. Second, we combine an edge-based force derived from the proposed edge detector with a new region-based force driven by the Bhattacharyya gradient flow and adapted to the case of parametric active surfaces. Finally, we develop a quasi-automatic initialization technique for deformable models by analyzing the profiles of the proposed edge detector response radially to obtain initial landmark points toward which an initial surface model is warped. RESULTS: We validate our method on a set of 36 TRUS images for which manual delineations were performed by two expert radiation oncologists, using a wide variety of quantitative metrics. The proposed hybrid model achieved state-of-the-art segmentation accuracy. CONCLUSION: Results demonstrate the interest of the proposed hybrid framework for accurate prostate volume segmentation. SIGNIFICANCE: This paper presents a modular framework for accurate prostate volume segmentation in TRUS, broadening the range of available strategies to tackle this open problem.

Tumour functional sphericity from PET images: prognostic value in NSCLC and impact of delineation method.

PURPOSE: Sphericity has been proposed as a parameter for characterizing PET tumour volumes, with complementary prognostic value with respect to SUV and volume in both head and neck cancer and lung cancer. The objective of the present study was to investigate its dependency on tumour delineation and the resulting impact on its prognostic value. METHODS: Five segmentation methods were considered: two thresholds (40% and 50% of SUVmax), ant colony optimization, fuzzy locally adaptive Bayesian (FLAB), and gradient-aided region-based active contour. The accuracy of each method in extracting sphericity was evaluated using a dataset of 176 simulated, phantom and clinical PET images of tumours with associated ground truth. The prognostic value of sphericity and its complementary value with respect to volume for each segmentation method was evaluated in a cohort of 87 patients with stage II/III lung cancer. RESULTS: Volume and associated sphericity values were dependent on the segmentation method. The correlation between segmentation accuracy and sphericity error was moderate (|ρ| from 0.24 to 0.57). The accuracy in measuring sphericity was not dependent on volume (|ρ| < 0.4). In the patients with lung cancer, sphericity had prognostic value, although lower than that of volume, except for that derived using FLAB for which when combined with volume showed a small improvement over volume alone (hazard ratio 2.67, compared with 2.5). Substantial differences in patient prognosis stratification were observed depending on the segmentation method used. CONCLUSION: Tumour functional sphericity was found to be dependent on the segmentation method, although the accuracy in retrieving the true sphericity was not dependent on tumour volume. In addition, even accurate segmentation can lead to an inaccurate sphericity value, and vice versa. Sphericity had similar or lower prognostic value than volume alone in the patients with lung cancer, except when determined using the FLAB method for which there was a small improvement in stratification when the parameters were combined.

The first MICCAI challenge on PET tumor segmentation.

INTRODUCTION: Automatic functional volume segmentation in PET images is a challenge that has been addressed using a large array of methods. A major limitation for the field has been the lack of a benchmark dataset that would allow direct comparison of the results in the various publications. In the present work, we describe a comparison of recent methods on a large dataset following recommendations by the American Association of Physicists in Medicine (AAPM) task group (TG) 211, which was carried out within a MICCAI (Medical Image Computing and Computer Assisted Intervention) challenge. MATERIALS AND METHODS: Organization and funding was provided by France Life Imaging (FLI). A dataset of 176 images combining simulated, phantom and clinical images was assembled. A website allowed the participants to register and download training data (n = 19). Challengers then submitted encapsulated pipelines on an online platform that autonomously ran the algorithms on the testing data (n = 157) and evaluated the results. The methods were ranked according to the arithmetic mean of sensitivity and positive predictive value. RESULTS: Sixteen teams registered but only four provided manuscripts and pipeline(s) for a total of 10 methods. In addition, results using two thresholds and the Fuzzy Locally Adaptive Bayesian (FLAB) were generated. All competing methods except one performed with median accuracy above 0.8. The method with the highest score was the convolutional neural network-based segmentation, which significantly outperformed 9 out of 12 of the other methods, but not the improved K-Means, Gaussian Model Mixture and Fuzzy C-Means methods. CONCLUSION: The most rigorous comparative study of PET segmentation algorithms to date was carried out using a dataset that is the largest used in such studies so far. The hierarchy amongst the methods in terms of accuracy did not depend strongly on the subset of datasets or the metrics (or combination of metrics). All the methods submitted by the challengers except one demonstrated good performance with median accuracy scores above 0.8.