The editorial introduces the Special Section on Radiogenomics in Prognosis and Treatment for Volume 8 Issue 3 of the Journal of Medical Imaging.

The power of predictive modeling for radiotherapy outcomes has historically been limited by an inability to adequately capture patient-specific variabilities; however, next-generation platforms together with imaging technologies and powerful bioinformatic tools have facilitated strategies and provided optimism. Integrating clinical, biological, imaging, and treatment-specific data for more accurate prediction of tumor control probabilities or risk of radiation-induced side effects are high-dimensional problems whose solutions could have widespread benefits to a diverse patient population—we discuss technical approaches toward this objective. Increasing interest in the above is specifically reflected by the emergence of two nascent fields, which are distinct but complementary: radiogenomics, which broadly seeks to integrate biological risk factors together with treatment and diagnostic information to generate individualized patient risk profiles, and radiomics, which further leverages large-scale imaging correlates and extracted features for the same purpose. We review classical analytical and data-driven approaches for outcomes prediction that serve as antecedents to both radiomic and radiogenomic strategies. Discussion then focuses on uses of conventional and deep machine learning in radiomics. We further consider promising strategies for the harmonization of high-dimensional, heterogeneous multiomics datasets (panomics) and techniques for nonparametric validation of best-fit models. Strategies to overcome common pitfalls that are unique to data-intensive radiomics are also discussed.

Purpose: The purpose of our study was to combine differences in radiomic features extracted from lung regions in the computed tomography (CT) scans of patients diagnosed with idiopathic pulmonary fibrosis (IPF) to identify associations with genetic variations and patient survival.

Approach: A database of CT scans and genomic data from 169 patients diagnosed with IPF was collected retrospectively. Six region-of-interest pairs (three per lung, positioned posteriorly, anteriorly, and laterally) were placed in each of three axial CT sections for each patient. Thirty-one features were used in logistic regression to classify patients’ genetic mutation status; classification performance was evaluated through the area under the receiver operating characteristic (ROC) curve [average area under the ROC curve (AUC)]. Kaplan–Meier (KM) survival curve models quantified the ability of each feature to differentiate between survival curves based on feature-specific thresholds.

Results: Nine first-order texture features and one fractal feature were correlated with TOLLIP-1 (rs4963062) mutations (AUC: 0.54 to 0.74), and five Laws’ filter features were correlated with TOLLIP-2 (rs5743905) mutations (AUC: 0.53 to 0.70). None of the features analyzed were found to be correlated with MUC5B mutations. First-order and fractal features demonstrated the greatest discrimination between KM curves.

Conclusions: A radiomics approach for the correlation of patient genetic mutations with image texture features has potential as a biomarker. These features also may serve as prognostic indicators using a survival curve modeling approach in which the combination of radiomic features and genetic mutations provides an enhanced understanding of the interaction between imaging phenotype and patient genotype on the progression and treatment of IPF.

Purpose: The goal of this study is to develop innovative methods for identifying radiomic features that are reproducible over varying image acquisition settings.

Approach: We propose a regularized partial correlation network to identify reliable and reproducible radiomic features. This approach was tested on two radiomic feature sets generated using two different reconstruction methods on computed tomography (CT) scans from a cohort of 47 lung cancer patients. The largest common network component between the two networks was tested on phantom data consisting of five cancer samples. To further investigate whether radiomic features found can identify phenotypes, we propose a k-means clustering algorithm coupled with the optimal mass transport theory. This approach following the regularized partial correlation network analysis was tested on CT scans from 77 head and neck squamous cell carcinoma (HNSCC) patients in the Cancer Imaging Archive (TCIA) and validated using an independent dataset.

Results: A set of common radiomic features was found in relatively large network components between the resultant two partial correlation networks resulting from a cohort of lung cancer patients. The reliability and reproducibility of those radiomic features were further validated on phantom data using the Wasserstein distance. Further analysis using the network-based Wasserstein k-means algorithm on the TCIA HNSCC data showed that the resulting clusters separate tumor subsites as well as HPV status, and this was validated on an independent dataset.

Conclusion: We showed that a network-based analysis enables identifying reproducible radiomic features and use of the selected set of features can enhance clustering results.

Purpose: Radiogenomics offers a potential virtual and noninvasive biopsy. However, radiogenomics models often suffer from generalizability issues, which cause a performance degradation on unseen data. In MRI, differences in the sequence parameters, manufacturers, and scanners make this generalizability issue worse. Such image acquisition information may be used to define different environments and select robust and invariant radiomic features associated with the clinical outcome that should be included in radiomics/radiogenomics models.

Approach: We assessed 77 low-grade gliomas and glioblastomas multiform patients publicly available in TCGA and TCIA. Radiomics features were extracted from multiparametric MRI images (T1-weighted, contrast-enhanced T1-weighted, T2-weighted, and fluid-attenuated inversion recovery) and different regions-of-interest (enhancing tumor, nonenhancing tumor/necrosis, and edema). A method developed to find variables that are part of causal structures was used for feature selection and compared with an embedded feature selection approach commonly used in radiomics/radiogenomics studies, across two different scenarios: (1) leaving data from a center as an independent held-out test set and tuning the model with the data from the remaining centers and (2) use stratified partitioning to obtain the training and the held-out test sets.

Results: In scenario (1), the performance of the proposed methodology and the traditional embedded method was AUC: 0.75 [0.25; 1.00] versus 0.83 [0.50; 1.00], Sens.: 0.67 [0.20; 0.93] versus 0.67 [0.20; 0.93], Spec.: 0.75 [0.30; 0.95] versus 0.75 [0.30; 0.95], and MCC: 0.42 [0.19; 0.68] versus 0.42 [0.19; 0.68] for center 1 as the held-out test set. The performance of both methods for center 2 as the held-out test set was AUC: 0.64 [0.36; 0.91] versus 0.55 [0.27; 0.82], Sens.: 0.00 [0.00; 0.73] versus 0.00 [0.00; 0.73], Spec.: 0.82 [0.52; 0.94] versus 0.91 [0.62; 0.98], and MCC: −0.13   [    −  0.38  ;    −  0.04  ]   versus −0.09   [    −  0.38  ;    −  0.02  ]  , whereas for center 3 was AUC: 0.80 [0.62; 0.95] versus 0.89 [0.56; 0.96], Sens.: 0.86 [0.48; 0.97] versus 0.86 [0.48; 0.97], Spec.: 0.72 [0.54; 0.85] versus 0.79 [0.61; 0.90], and MCC: 0.47 [0.41; 0.53] versus 0.55 [0.48; 0.60]. For center 4, the performance of both methods was AUC: 0.77 [0.51; 1.00] versus 0.75 [0.47; 0.97], Sens.: 0.53 [0.30; 0.75] versus 0.00 [0.00; 0.15], Spec.: 0.71 [0.35; 0.91] versus 0.86 [0.48; 0.97], and MCC: 0.23 [0.16; 0.31] versus. −0.32   [    −  0.46  ;    −  0.20  ]  . In scenario (2), the performance of these methods was AUC: 0.89 [0.71; 1.00] versus 0.79 [0.58; 0.94], Sens.: 0.86 [0.80; 0.92] versus 0.43 [0.15; 0.74], Spec.: 0.87 [0.62; 0.96] versus 0.87 [0.62; 0.96], and MCC: 0.70 [0.60; 0.77] versus 0.33 [0.24; 0.42].

Conclusions: This proof-of-concept study demonstrated good performance by the proposed feature selection method in the majority of the studied scenarios, as it promotes robustness of features included in the models and the models’ generalizability by making used imaging data of different scanners or with sequence parameters.

Purpose: Integrative analysis combining diagnostic imaging and genomic information can uncover biological insights into lesions that are visible on radiologic images. We investigate techniques for interrogating a deep neural network trained to predict quantitative image (radiomic) features and histology from gene expression in non-small cell lung cancer (NSCLC).

Approach: Using 262 training and 89 testing cases from two public datasets, deep feedforward neural networks were trained to predict the values of 101 computed tomography (CT) radiomic features and histology. A model interrogation method called gene masking was used to derive the learned associations between subsets of genes and a radiomic feature or histology class [adenocarcinoma (ADC), squamous cell, and other].

Results: Overall, neural networks outperformed other classifiers. In testing, neural networks classified histology with area under the receiver operating characteristic curves (AUCs) of 0.86 (ADC), 0.91 (squamous cell), and 0.71 (other). Classification performance of radiomics features ranged from 0.42 to 0.89 AUC. Gene masking analysis revealed new and previously reported associations. For example, hypoxia genes predicted histology (>0.90  AUC). Previously published gene signatures for classifying histology were also predictive in our model (>0.80  AUC). Gene sets related to the immune or cardiac systems and cell development processes were predictive (>0.70  AUC) of several different radiomic features. AKT signaling, tumor necrosis factor, and Rho gene sets were each predictive of tumor textures.

Conclusions: This work demonstrates neural networks’ ability to map gene expressions to radiomic features and histology types in NSCLC and to interpret the models to identify predictive genes associated with each feature or type.

The field of radiogenomics largely focuses on developing imaging surrogates for genomic signatures and integrating imaging, genomic, and molecular data to develop combined personalized biomarkers for characterizing various diseases. Our study aims to highlight the current state-of-the-art and the role of radiogenomics in cancer research, focusing mainly on solid tumors, and is broadly divided into four sections. The first section reviews representative studies that establish the biologic basis of radiomic signatures using gene expression and molecular profiling information. The second section includes studies that aim to non-invasively predict molecular subtypes of tumors using radiomic signatures. The third section reviews studies that evaluate the potential to augment the performance of established prognostic signatures by combining complementary information encoded by radiomic and genomic signatures derived from cancer tumors. The fourth section includes studies that focus on ascertaining the biological significance of radiomic phenotypes. We conclude by discussing current challenges and opportunities in the field, such as the importance of coordination between imaging device manufacturers, regulatory organizations, health care providers, pharmaceutical companies, academic institutions, and physicians for the effective standardization of the results from radiogenomic signatures and for the potential use of these findings to improve precision care for cancer patients.

Purpose: Deep convolutional neural networks (CNN) have demonstrated impressive success in various image classification tasks. We investigated the use of CNNs to distinguish between benign and malignant microcalcifications, using either conventional or dual-energy mammography x-ray images. The two kinds of calcifications, known as type-I (calcium oxalate crystals) and type-II (calcium phosphate aggregations), have different attenuation properties in the mammographic energy range. However, variations in microcalcification shape, size, and density as well as compressed breast thickness and breast tissue background make this a challenging discrimination task for the human visual system.

Approach: Simulations (conventional and dual-energy mammography) and phantom experiments (conventional mammography only) were conducted using the range of breast thicknesses and randomly shaped microcalcifications. The off-the-shelf Resnet-18 CNN was trained on the regions of interest with calcification clusters of the two kinds.

Results: Both Monte Carlo simulations and experimental phantom data suggest that deep neural networks can be trained to separate the two classes of calcifications with high accuracy, using dual-energy mammograms.

Conclusions: Our work shows the encouraging results of using the CNNs for non-invasive testing for type-I and type-II microcalcifications and may stimulate further research in this area with expanding presence of the novel breast imaging modalities like dual-energy mammography or systems using photon-counting detectors.

Purpose: To validate a previously proposed algorithm that modifies a mammogram to appear as if it was acquired with different technique factors using realistic phantom-based mammograms.

Approach: Two digital mammography systems (an indirect- and a direct-detector-based system) were used to acquire realistic mammographic images of five 3D-printed breast phantoms with the technique factors selected by the automatic exposure control and at various other conditions (denoted by the original images). Additional images under other simulated conditions were also acquired: higher or lower tube voltages, different anode/filter combinations, or lower tube current–time products (target images). The signal and noise in the original images were modified to simulate the target images (simulated images). The accuracy of the image modification algorithm was validated by comparing the target and simulated images using the local mean, local standard deviation (SD), local variance, and power spectra (PS) of the image signals. The absolute relative percent error between the target and simulated images for each parameter was calculated at each sub-region of interest (local parameters) and frequency (PS), and then averaged.

Results: The local mean signal, local SD, local variance, and PS of the target and simulated images were very similar, with a relative percent error of 5.5%, 3.8%, 7.8%, and 4.4% (indirect system), respectively, and of 3.7%, 3.8%, 7.7%, and 7.5% (direct system), respectively.

Conclusions: The algorithm is appropriate for simulating different technique factors. Therefore, it can be used in various studies, for instance to evaluate the impact of technique factors in cancer detection using clinical images.

Purpose: Physiological or pathological variation would cause a change of conductivity. Electrical impedance tomography (EIT) is favorable in reconstructing conductivity distribution inside the detected area. However, the reconstruction is an ill-posed inverse problem and the spatial resolution of the reconstructed image is relatively poor.

Approach: To deal with the problem, a regularization method is commonly applied. Traditional regularization methods have their own disadvantages. In this work, we develop an innovative hybrid regularization method to determine the conductivity distribution from the boundary measurement. To address the unwanted artifact observed in the total variation (TV) method, the proposed approach incorporates the TV method with the non-convex sparse penalty term-based wavelet transform. In the reconstruction, the sensitivity matrix is also normalized to increase the sensitivity of the measurement to the variation of the conductivity. The objective function is minimized with the split augmented Lagrangian shrinkage algorithm.

Results: The feasibility of the proposed method is evaluated by numerical simulation and phantom experiment. The results verify that the reconstruction with the proposed method is more advantageous, as obvious improvement is observed in the reconstructed image.

Conclusions: With the proposed method, the artifact can be effectively suppressed and the reconstructed image of conductivity distribution is improved. It has great potential in medical imaging, which would be helpful for the accurate diagnosis of disease.

Purpose: The purpose of this study was to evaluate the potential of a prototype gallium arsenide (GaAs) photon-counting detector (PCD) for imaging of the breast.

Approach: First, the contrast-to-noise ratio (CNR) using different aluminum/poly(methyl methacrylate) (PMMA) phantoms of different thicknesses were measured. Second, microcalcification detection accuracy using a receiver operating characteristic study with three observers reading an ensemble of images was measured. Finally, the feasibility of using a GaAs system with two energy bins for contrast-enhanced mammography was investigated.

Results: For the first two studies, the GaAs detector was compared with a commercial mammography system. The CNR was estimated by imaging 18-, 36-, and 110-μm-thick aluminum targets placed on top of 6 cm of PMMA plates and was found to be similar or better over a range of exposures. We observed a similar performance of detecting microcalcifications with the GaAs detector over a range of clinically applicable dose levels with a small increase at lower dose levels. The results also showed that contrast-enhanced spectral mammography using a GaAs PCD is feasible and beneficial.

Conclusions: Results from this study suggest that performance with this new detector seems either slightly improved or equivalent to a commercial mammography system that used an energy-integrated detector. No attempt at optimizing exposure techniques for the GaAs detector was performed. Further research is needed to determine optimal acquisition parameters for the GaAs detector and to develop more sophisticated material decomposition algorithms that promise to provide improved quantitative estimates of iodine uptake.

Purpose: The lack of standardization in quantitative radiomic measures of tumors seen on computed tomography (CT) scans is generally recognized as an unresolved issue. To develop reliable clinical applications, radiomics must be robust across different CT scan modes, protocols, software, and systems. We demonstrate how custom-designed phantoms, imprinted with human-derived patterns, can provide a straightforward approach to validating longitudinally stable radiomic signature values in a clinical setting.

Approach: Described herein is a prototype process to design an anatomically informed 3D-printed radiomic phantom. We used a multimaterial, ultra-high-resolution 3D printer with voxel printing capabilities. Multiple tissue regions of interest (ROIs), from four pancreas tumors, one lung tumor, and a liver background, were extracted from digital imaging and communication in medicine (DICOM) CT exam files and were merged together to develop a multipurpose, circular radiomic phantom (18 cm diameter and 4 cm width). The phantom was scanned 30 times using standard clinical CT protocols to test repeatability. Features that have been found to be prognostic for various diseases were then investigated for their repeatability and reproducibility across different CT scan modes.

Results: The structural similarity index between the segment used from the patients’ DICOM image and the phantom CT scan was 0.71. The coefficient variation for all assessed radiomic features was <1.0  %   across 30 repeat scans of the phantom. The percent deviation (pDV) from the baseline value, which was the mean feature value determined from repeat scans, increased with the application of the lung convolution kernel, changes to the voxel size, and increases in the image noise. Gray level co-occurrence features, contrast, dissimilarity, and entropy were particularly affected by different scan modes, presenting with pDV  >    ±  15  %  .

Conclusions: Previously discovered prognostic and popular radiomic features are variable in practice and need to be interpreted with caution or excluded from clinical implementation. Voxel-based 3D printing can reproduce tissue morphology seen on CT exams. We believe that this is a flexible, yet practical, way to design custom phantoms to validate and compare radiomic metrics longitudinally, over time, and across systems.

Purpose: Conventional automated segmentation of the head anatomy in magnetic resonance images distinguishes different brain and nonbrain tissues based on image intensities and prior tissue probability maps (TPMs). This works well for normal head anatomies but fails in the presence of unexpected lesions. Deep convolutional neural networks (CNNs) leverage instead spatial patterns and can learn to segment lesions but often ignore prior probabilities.

Approach: We add three sources of prior information to a three-dimensional (3D) convolutional network, namely, spatial priors with a TPM, morphological priors with conditional random fields, and spatial context with a wider field-of-view at lower resolution. We train and test these networks on 3D images of 43 stroke patients and 4 healthy individuals which have been manually segmented.

Results: We demonstrate the benefits of each source of prior information, and we show that the new architecture, which we call Multiprior network, improves the performance of existing segmentation software, such as SPM, FSL, and DeepMedic for abnormal anatomies. The relevance of the different priors was compared, and the TPM was found to be most beneficial. The benefit of adding a TPM is generic in that it can boost the performance of established segmentation networks such as the DeepMedic and a UNet. We also provide an out-of-sample validation and clinical application of the approach on an additional 47 patients with disorders of consciousness. We make the code and trained networks freely available.

Conclusions: Biomedical images follow imaging protocols that can be leveraged as prior information into deep CNNs to improve performance. The network segmentations match human manual corrections performed in 3D and are comparable in performance to human segmentations obtained from scratch in 2D for abnormal brain anatomies.

Purpose: Echocardiography is the most commonly used modality for assessing the heart in clinical practice. In an echocardiographic exam, an ultrasound probe samples the heart from different orientations and positions, thereby creating different viewpoints for assessing the cardiac function. The determination of the probe viewpoint forms an essential step in automatic echocardiographic image analysis.

Approach: In this study, convolutional neural networks are used for the automated identification of 14 different anatomical echocardiographic views (larger than any previous study) in a dataset of 8732 videos acquired from 374 patients. Differentiable architecture search approach was utilized to design small neural network architectures for rapid inference while maintaining high accuracy. The impact of the image quality and resolution, size of the training dataset, and number of echocardiographic view classes on the efficacy of the models were also investigated.

Results: In contrast to the deeper classification architectures, the proposed models had significantly lower number of trainable parameters (up to 99.9% reduction), achieved comparable classification performance (accuracy 88.4% to 96%, precision 87.8% to 95.2%, recall 87.1% to 95.1%) and real-time performance with inference time per image of 3.6 to 12.6 ms.

Conclusion: Compared with the standard classification neural network architectures, the proposed models are faster and achieve comparable classification performance. They also require less training data. Such models can be used for real-time detection of the standard views.

Purpose: Semi-automatic image segmentation is still a valuable tool in clinical applications since it retains the expert oversights legally required. However, semi-automatic methods for simultaneous multi-class segmentation are difficult to be clinically implemented due to the complexity of underlining algorithms. We purpose an efficient one-vs-rest graph cut approach of which the complexity only grows linearly as the number of classes increases.

Approach: Given an image slice, we construct multiple one-vs-rest graphs, each for a tissue class, for inference of a conditional random field (CRF). The one-vs-rest graph cut is to minimize the CRF energy derived from regional and boundary class probabilities estimated from random forests to obtain a one-vs-rest segmentation. The final segmentation is obtained by fusing from those one-vs-rest segmentations based on majority voting. We compare our method to a well-used multi-class graph cut method, alpha-beta swap, and a fully connected CRF (FCCRF) method, in brain tumor segmentation of 20 high-grade tumor cases in 2013 MICCAI dataset.

Results: Our method achieved mean Dice score of 0.83 for whole tumor, compared to 0.80 by alpha-beta swap and 0.79 by FCCRF. There was a performance improvement over alpha-beta swap by a factor of five.

Conclusions: Our method utilizes the probabilistic-based CRF which can be estimated from any machine learning technique. Comparing to traditional multi-class graph cut, the purposed one-vs-rest approach has complexity that grows only linearly as the number of classes increases, therefore, our method can be applicable for both online semi-automatic and offline automatic segmentation in clinical applications.

Purpose: In-hospital three-dimensional (3D) printing of patient-specific pathologies is increasingly being used in daily care. However, the efficiency of the current conversion from image to print is often obstructed due to limitations associated with segmentation software. Therefore, there is a need for comparison of several clinically available tools. A comparative study has been conducted to compare segmentation performance of Philips IntelliSpace Portal^® (PISP), Mimics Innovation Suite (MIS), and DICOM to PRINT^® (D2P).

Approach: These tools were compared with respect to segmentation time and 3D mesh quality. The dataset consisted of three computed tomography (CT)-scans of acetabular fractures (ACs), three CT-scans of tibia plateau fractures (TPs), and three CTA-scans of abdominal aortic aneurysms (AAAs). Independent-samples t-tests were performed to compare the measured segmentation times. Furthermore, 3D mesh quality was assessed and compared according to representativeness and usability for the surgeon.

Results: Statistically significant differences in segmentation time were found between PISP and MIS with respect to the segmentation of ACs (p  =    <  0.001) and AAAs (p  =  0.031). Furthermore, statistically significant differences in segmentation time were found between PISP and D2P for segmentations of AAAs (p  =  0.008). There were no statistically significant differences in segmentation time for TPs. The accumulated mesh quality scores were highest for segmentations performed in MIS, followed by D2P.

Conclusion: Based on segmentation time and mesh quality, MIS and D2P are capable of enhancing the in-hospital 3D print workflow. However, they should be integrated with the picture archiving and communication system to truly improve the workflow. In addition, these software packages are not open source and additional costs must be incurred.

Purpose: The Breast Pathology Quantitative Biomarkers (BreastPathQ) Challenge was a Grand Challenge organized jointly by the international society for optics and photonics (SPIE), the American Association of Physicists in Medicine (AAPM), the U.S. National Cancer Institute (NCI), and the U.S. Food and Drug Administration (FDA). The task of the BreastPathQ Challenge was computerized estimation of tumor cellularity (TC) in breast cancer histology images following neoadjuvant treatment.

Approach: A total of 39 teams developed, validated, and tested their TC estimation algorithms during the challenge. The training, validation, and testing sets consisted of 2394, 185, and 1119 image patches originating from 63, 6, and 27 scanned pathology slides from 33, 4, and 18 patients, respectively. The summary performance metric used for comparing and ranking algorithms was the average prediction probability concordance (PK) using scores from two pathologists as the TC reference standard.

Results: Test PK performance ranged from 0.497 to 0.941 across the 100 submitted algorithms. The submitted algorithms generally performed well in estimating TC, with high-performing algorithms obtaining comparable results to the average interrater PK of 0.927 from the two pathologists providing the reference TC scores.

Conclusions: The SPIE-AAPM-NCI BreastPathQ Challenge was a success, indicating that artificial intelligence/machine learning algorithms may be able to approach human performance for cellularity assessment and may have some utility in clinical practice for improving efficiency and reducing reader variability. The BreastPathQ Challenge can be accessed on the Grand Challenge website.

Purpose: A method for fluoroscopic guidance of a robotic assistant is presented for instrument placement in pelvic trauma surgery. The solution uses fluoroscopic images acquired in standard clinical workflow and helps avoid repeat fluoroscopy commonly performed during implant guidance.

Approach: Images acquired from a mobile C-arm are used to perform 3D–2D registration of both the patient (via patient CT) and the robot (via CAD model of a surgical instrument attached to its end effector, e.g; a drill guide), guiding the robot to target trajectories defined in the patient CT. The proposed approach avoids C-arm gantry motion, instead manipulating the robot to acquire disparate views of the instrument. Phantom and cadaver studies were performed to determine operating parameters and assess the accuracy of the proposed approach in aligning a standard drill guide instrument.

Results: The proposed approach achieved average drill guide tip placement accuracy of 1.57  ±  0.47  mm and angular alignment of 0.35  ±  0.32  deg in phantom studies. The errors remained within 2 mm and 1 deg in cadaver experiments, comparable to the margins of errors provided by surgical trackers (but operating without the need for external tracking).

Conclusions: By operating at a fixed fluoroscopic perspective and eliminating the need for encoded C-arm gantry movement, the proposed approach simplifies and expedites the registration of image-guided robotic assistants and can be used with simple, non-calibrated, non-encoded, and non-isocentric C-arm systems to accurately guide a robotic device in a manner that is compatible with the surgical workflow.

Purpose: In dental offices, there is a trend replacing conventional silicone impressions and plaster cast models by imaging data of intraoral scanners to map the denture and surrounding tissues. The aim of the study is the analysis of the accuracy of selected commercially available scanners. The accuracy is considered as the main drawback in comparison to the conventional approach.

Approach: We evaluated the reproduction performance of five optical scanners by a direct comparison with high-resolution hard x-ray computed tomography data, all obtained from a polyetheretherketone model with similarity to a full-arch upper jaw.

Results: Using the software GOM Inspect (GOM GmbH, Braunschweig, Germany), we could classify the intraoral scanners into two groups. The more accurate instruments gave rise to the following precision values: 35  μm (TRIOS^® 3, 3shape, Copenhagen, Denmark), 43  μm (CS 3600, Carestream, Atlanta, Georgia), and 46  μm (3M™ True Definition Scanner, 3M ESPE, St. Paul, Minnesota). The less precise systems yielded 93  μm (Medit i500, Medit corp., Seongbuk-gu, South Korea) and 97  μm (Emerald™, Planmeca Oy, Helsinki, Finland).

Conclusions: The selected scanners are suitable for single crowns, small bridges, and separate quadrants prostheses. Scanners based on triangulation are hardly appropriate for full-arch prostheses. Besides precision, however, the choice of the scanner depends on scanning time, intraoral-camera size, and the user’s learning curve. The developed protocol, which includes three-dimensional (3D) imaging and advanced computational tools for the registration with the design data, will be increasingly used in geometrical metrology by nondestructive procedures to perform dimensional measurements with micrometer precision and is capable for detailed 3D geometrical models reconstruction.

Purpose: We investigate the impact of various deep-learning-based methods for detecting and segmenting metastases with different lesion volume sizes on 3D brain MR images.

Approach: A 2.5D U-Net and a 3D U-Net were selected. We also evaluated weak learner fusion of the prediction features generated by the 2.5D and the 3D networks. A 3D fully convolutional one-stage (FCOS) detector was selected as a representative of bounding-box regression-based detection methods. A total of 422 3D post-contrast T1-weighted scans from patients with brain metastases were used. Performances were analyzed based on lesion volume, total metastatic volume per patient, and number of lesions per patient.

Results: The performance of detection of the 2.5D and 3D U-Net methods had recall of >0.83 and precision of >0.44 for lesion volume >0.3  cm³ but deteriorated as metastasis size decreased below 0.3  cm³ to 0.58 to 0.74 in recall and 0.16 to 0.25 in precision. Compared the two U-Nets for detection capability, high precision was achieved by the 2.5D network, but high recall was achieved by the 3D network for all lesion sizes. The weak learner fusion achieved a balanced performance between the 2.5D and 3D U-Nets; particularly, it increased precision to 0.83 for lesion volumes of 0.1 to 0.3  cm³ but decreased recall to 0.59. The 3D FCOS detector did not outperform the U-Net methods in detecting either the small or large metastases presumably because of the limited data size.

Conclusions: Our study provides the performances of four deep learning methods in relationship to lesion size, total metastasis volume, and number of lesions per patient, providing insight into further development of the deep learning networks.