Clinical consequences of computer-aided colorectal polyp detection.

BACKGROUND AND AIM: Randomised trials show improved polyp detection with computer-aided detection (CADe), mostly of small lesions. However, operator and selection bias may affect CADe's true benefit. Clinical outcomes of increased detection have not yet been fully elucidated. METHODS: In this multicentre trial, CADe combining convolutional and recurrent neural networks was used for polyp detection. Blinded endoscopists were monitored in real time by a second observer with CADe access. CADe detections prompted reinspection. Adenoma detection rates (ADR) and polyp detection rates were measured prestudy and poststudy. Histological assessments were done by independent histopathologists. The primary outcome compared polyp detection between endoscopists and CADe. RESULTS: In 946 patients (51.9% male, mean age 64), a total of 2141 polyps were identified, including 989 adenomas. CADe was not superior to human polyp detection (sensitivity 94.6% vs 96.0%) but outperformed them when restricted to adenomas. Unblinding led to an additional yield of 86 true positive polyp detections (1.1% ADR increase per patient; 73.8% were <5 mm). CADe also increased non-neoplastic polyp detection by an absolute value of 4.9% of the cases (1.8% increase of entire polyp load). Procedure time increased with 6.6±6.5 min (+42.6%). In 22/946 patients, the additional detection of adenomas changed surveillance intervals (2.3%), mostly by increasing the number of small adenomas beyond the cut-off. CONCLUSION: Even if CADe appears to be slightly more sensitive than human endoscopists, the additional gain in ADR was minimal and follow-up intervals rarely changed. Additional inspection of non-neoplastic lesions was increased, adding to the inspection and/or polypectomy workload.

Lessons for future clinical trials in adults with Becker muscular dystrophy: Disease progression detected by muscle magnetic resonance imaging, clinical and patient-reported outcome measures.

BACKGROUND AND PURPOSE: Because Becker muscular dystrophy (BMD) is a heterogeneous disease and only few studies have evaluated adult patients, it is currently still unclear which outcome measures should be used in future clinical trials. METHODS: Muscle magnetic resonance imaging, patient-reported outcome measures and a wide range of clinical outcome measures, including motor function, muscle strength and timed-function tests, were evaluated in 21 adults with BMD at baseline and at 9 and 18 months of follow-up. RESULTS: Proton density fat fraction increased significantly in 10/17 thigh muscles after 9 months, and in all thigh and lower leg muscles after 18 months. The 32-item Motor Function Measurement (MFM-32) scale (-1.3%, p = 0.017), North Star Ambulatory Assessment (-1.3 points, p = 0.010) and patient-reported activity limitations scale (-0.3 logits, p = 0.018) deteriorated significantly after 9 months. The 6-min walk distance (-28.7 m, p = 0.042), 10-m walking test (-0.1 m/s, p = 0.032), time to climb four stairs test (-0.03 m/s, p = 0.028) and Biodex peak torque measurements of quadriceps (-4.6 N m, p = 0.014) and hamstrings (-5.0 N m, p = 0.019) additionally deteriorated significantly after 18 months. At this timepoint, domain 1 of the MFM-32 was the only clinical outcome measure with a large sensitivity to change (standardized response mean 1.15). DISCUSSION: It is concluded that proton density fat fraction imaging of entire thigh muscles is a sensitive outcome measure to track progressive muscle fat replacement in patients with BMD, already after 9 months of follow-up. Finally, significant changes are reported in a wide range of clinical and patient-reported outcome measures, of which the MFM-32 appeared to be the most sensitive to change in adults with BMD.

Deep learning pipeline for quality filtering of MRSI spectra.

With the rise of novel 3D magnetic resonance spectroscopy imaging (MRSI) acquisition protocols in clinical practice, which are capable of capturing a large number of spectra from a subject's brain, there is a need for an automated preprocessing pipeline that filters out bad-quality spectra and identifies contaminated but salvageable spectra prior to the metabolite quantification step. This work introduces such a pipeline based on an ensemble of deep-learning classifiers. The dataset consists of 36,338 spectra from one healthy subject and five brain tumor patients, acquired with an EPSI variant, which implemented a novel type of spectral editing named SLOtboom-Weng (SLOW) editing on a 7T MR scanner. The spectra were labeled manually by an expert into four classes of spectral quality as follows: (i) noise, (ii) spectra greatly influenced by lipid-related artifacts (deemed not to contain clinical information), (iii) spectra containing metabolic information slightly contaminated by lipid signals, and (iv) good-quality spectra. The AI model consists of three pairs of networks, each comprising a convolutional autoencoder and a multilayer perceptron network. In the classification step, the encoding half of the autoencoder is kept as a dimensionality reduction tool, while the fully connected layers are added to its output. Each of the three pairs of networks is trained on different representations of spectra (real, imaginary, or both), aiming at robust decision-making. The final class is assigned via a majority voting scheme. The F1 scores obtained on the test dataset for the four previously defined classes are 0.96, 0.93, 0.82, and 0.90, respectively. The arguably lower value of 0.82 was reached for the least represented class of spectra mildly influenced by lipids. Not only does the proposed model minimise the required user interaction, but it also greatly reduces the computation time at the metabolite quantification step (by selecting a subset of spectra worth quantifying) and enforces the display of only clinically relevant information.

Advanced Imaging in Gastrointestinal Endoscopy: A Literature Review of the Current State of the Art.

Background and Aims: Gastrointestinal (GI) endoscopy has known a great evolution in the last decades. Imaging techniques evolved from imaging with only standard white light endoscopes toward high-definition resolution endoscopes and the use of multiple color enhancement techniques, over to automated endoscopic assessment systems based on artificial intelligence. This narrative literature review aimed to provide a detailed overview on the latest evolutions within the field of advanced GI endoscopy, mainly focusing on the screening, diagnosis, and surveillance of common upper and lower GI pathology. Methods: This review comprises only literature about screening, diagnosis, and surveillance strategies using advanced endoscopic imaging techniques published in (inter)national peer-reviewed journals and written in English. Studies with only adult patients included were selected. A search was performed using MESH terms: dye-based chromoendoscopy, virtual chromoendoscopy, video enhancement technique, upper GI tract, lower GI tract, Barrett's esophagus, esophageal squamous cell carcinoma, gastric cancer, colorectal polyps, inflammatory bowel disease, artificial intelligence. This review does not elaborate on the therapeutic application or impact of advanced GI endoscopy. Conclusions: Focusing on current and future applications and evolutions in the field of both upper and lower GI advanced endoscopy, this overview is a practical but detailed projection of the latest developments. Within this review, an active leap toward artificial intelligence and its recent developments in GI endoscopy was made. Additionally, the literature is weighted against the current international guidelines and assessed for its potential positive future impact.

Introdução/objetivos: A endoscopia digestiva conheceu uma grande evolução nas últimas décadas, tendo as técnicas de imagem evoluído de imagens com luz branca para endoscópios de alta definição com possibilidade de uso de várias técnicas de melhoramento de cores e até sistemas automatizados apoiados em inteligência artificial. Esta revisão narrativa da literatura visa fornecer uma visão detalhada das últimas evoluções no campo da endoscopia avançada, focando principalmente no rastreio, diagnóstico e vigilância. Métodos: Pesquisa da literatura sobre estratégias de rastreio, diagnóstico e vigilância utilizando técnicas avançadas de imagem endoscópica publicadas em revistas internacionais revistas por pares e escritas em inglês. Foram selecionados estudos apenas com doentes adultos e foi realizada pesquisa utilizando termos MESH: cromoendoscopia com corante, cromoendoscopia virtual, técnicas de melhoramento de vídeo, tubo digestivo superior, tubo digestivo inferior, esófago de Barrett, carcinoma de células escamosas, cancro gástrico, pólipos colorretais, doença inflamatória intestinal e inteligência artificial. Conclusões: Esta revisão avaliou de uma forma prática os últimos desenvolvimentos no campo da imagem avançada em endoscopia digestiva, avaliando-se também as perspetivas futuras e o potencial impacto da inteligência artificial.

Automated MRI quantification of volumetric per-muscle fat fractions in the proximal leg of patients with muscular dystrophies.

Muscular dystrophies (MD) are a class of rare genetic diseases resulting in progressive muscle weakness affecting specific muscle groups, depending on the type of disease. Disease progression is characterized by the gradual replacement of muscle tissue by fat, which can be assessed with fat-sensitive magnetic resonance imaging (MRI) and objectively evaluated by quantifying the fat fraction percentage (FF%) per muscle. Volumetric quantification of fat replacement over the full 3D extent of each muscle is more precise and potentially more sensitive than 2D quantification in few selected slices only, but it requires an accurate 3D segmentation of each muscle individually, which is time consuming when this has to be performed manually for a large number of muscles. A reliable, largely automated approach for 3D muscle segmentation is thus needed to facilitate the adoption of fat fraction quantification as a measure of MD disease progression in clinical routine practice, but this is challenging due to the variable appearance of the images and the ambiguity in the discrimination of the contours of adjacent muscles, especially when the normal image contrast is affected and diminished by the fat replacement. To deal with these challenges, we used deep learning to train AI-models to segment the muscles in the proximal leg from knee to hip in Dixon MRI images of healthy subjects as well as patients with MD. We demonstrate state-of-the-art segmentation results of all 18 muscles individually in terms of overlap (Dice score, DSC) with the manual ground truth delineation for images of cases with low fat infiltration (mean overall FF%: 11.3%; mean DSC: 95.3% per image, 84.4-97.3% per muscle) as well as with medium and high fat infiltration (mean overall FF%: 44.3%; mean DSC: 89.0% per image, 70.8-94.5% per muscle). In addition, we demonstrate that the segmentation performance is largely invariant to the field of view of the MRI scan, is generalizable to patients with different types of MD and that the manual delineation effort to create the training set can be drastically reduced without significant loss of segmentation quality by delineating only a subset of the slices.

Histopathological correlations and fat replacement imaging patterns in recessive limb-girdle muscular dystrophy type 12.

BACKGROUND: Despite the widespread use of proton density fat fraction (PDFF) measurements with magnetic resonance imaging (MRI) to track disease progression in muscle disorders, it is still unclear how these findings relate to histopathological changes in muscle biopsies of patients with limb-girdle muscular dystrophy autosomal recessive type 12 (LGMDR12). Furthermore, although it is known that LGMDR12 leads to a selective muscle involvement distinct from other muscular dystrophies, the spatial distribution of fat replacement within these muscles is unknown. METHODS: We included 27 adult patients with LGMDR12 and 27 age-matched and sex-matched healthy controls and acquired 6-point Dixon images of the thighs and T1 and short tau inversion recovery (STIR) MR images of the whole body. In 16 patients and 15 controls, we performed three muscle biopsies, one in the semimembranosus, vastus lateralis, and rectus femoris muscles, which are severely, intermediately, and mildly affected in LGMDR12, respectively. We correlated the PDFF to the fat percentage measured on biopsies of the corresponding muscles, as well as to the Rochester histopathology grading scale. RESULTS: In patients, we demonstrated a strong correlation of PDFF on MRI and muscle biopsy fat percentage for the semimembranosus (r = 0.85, P < 0.001) and vastus lateralis (r = 0.68, P = 0.005). We found similar results for the correlation between PDFF and the Rochester histopathology grading scale. Out of the five patients with inflammatory changes on muscle biopsy, three showed STIR hyperintensities in the corresponding muscle on MRI. By modelling the PDFF on MRI for 18 thigh muscles from origin to insertion, we observed a significantly inhomogeneous proximo-distal distribution of fat replacement in all thigh muscles of patients with LGMDR12 (P < 0.001), and different patterns of fat replacement within each of the muscles. CONCLUSIONS: We showed a strong correlation of fat fraction on MRI and fat percentage on muscle biopsy for diseased muscles and validated the use of Dixon fat fraction imaging as an outcome measure in LGMDR12. The inhomogeneous fat replacement within thigh muscles on imaging underlines the risk of analysing only samples of muscles instead of the entire muscles, which has important implications for clinical trials.

Benefits of automated gross tumor volume segmentation in head and neck cancer using multi-modality information.

PURPOSE: Gross tumor volume (GTV) delineation for head and neck cancer (HNC) radiation therapy planning is time consuming and prone to interobserver variability (IOV). The aim of this study was (1) to develop an automated GTV delineation approach of primary tumor (GTVp) and pathologic lymph nodes (GTVn) based on a 3D convolutional neural network (CNN) exploiting multi-modality imaging input as required in clinical practice, and (2) to validate its accuracy, efficiency and IOV compared to manual delineation in a clinical setting. METHODS: Two datasets were retrospectively collected from 150 clinical cases. CNNs were trained for GTV delineation with consensus delineation as ground truth, with either single (CT) or co-registered multi-modal (CT + PET or CT + MRI) imaging data as input. For validation, GTVs were delineated on 20 new cases by two observers, once manually, once by correcting the delineations generated by the CNN. RESULTS: Both multi-modality CNNs performed better than the single-modality CNN and were selected for clinical validation. Mean Dice Similarity Coefficient (DSC) for (GTVp, GTVn) respectively between automated and manual delineations was (69%, 79%) for CT + PET and (59%,71%) for CT + MRI. Mean DSC between automated and corrected delineations was (81%,89%) for CT + PET and (69%,77%) for CT + MRI. Mean DSC between observers was (76%,86%) for manual delineations and (95%,96%) for corrected delineations, indicating a significant decrease in IOV (p < 10-5), while efficiency increased significantly (48%, p < 10-5). CONCLUSION: Multi-modality automated delineation of GTV of HNC was shown to be more efficient and consistent compared to manual delineation in a clinical setting and beneficial over a single-modality approach.

Factorizer: A scalable interpretable approach to context modeling for medical image segmentation.

Convolutional Neural Networks (CNNs) with U-shaped architectures have dominated medical image segmentation, which is crucial for various clinical purposes. However, the inherent locality of convolution makes CNNs fail to fully exploit global context, essential for better recognition of some structures, e.g., brain lesions. Transformers have recently proven promising performance on vision tasks, including semantic segmentation, mainly due to their capability of modeling long-range dependencies. Nevertheless, the quadratic complexity of attention makes existing Transformer-based models use self-attention layers only after somehow reducing the image resolution, which limits the ability to capture global contexts present at higher resolutions. Therefore, this work introduces a family of models, dubbed Factorizer, which leverages the power of low-rank matrix factorization for constructing an end-to-end segmentation model. Specifically, we propose a linearly scalable approach to context modeling, formulating Nonnegative Matrix Factorization (NMF) as a differentiable layer integrated into a U-shaped architecture. The shifted window technique is also utilized in combination with NMF to effectively aggregate local information. Factorizers compete favorably with CNNs and Transformers in terms of accuracy, scalability, and interpretability, achieving state-of-the-art results on the BraTS dataset for brain tumor segmentation and ISLES'22 dataset for stroke lesion segmentation. Highly meaningful NMF components give an additional interpretability advantage to Factorizers over CNNs and Transformers. Moreover, our ablation studies reveal a distinctive feature of Factorizers that enables a significant speed-up in inference for a trained Factorizer without any extra steps and without sacrificing much accuracy. The code and models are publicly available at https://github.com/pashtari/factorizer.

Classification of multiple sclerosis clinical profiles using machine learning and grey matter connectome.

Purpose: The main goal of this study is to investigate the discrimination power of Grey Matter (GM) thickness connectome data between Multiple Sclerosis (MS) clinical profiles using statistical and Machine Learning (ML) methods. Materials and Methods: A dataset composed of 90 MS patients acquired at the MS clinic of Lyon Neurological Hospital was used for the analysis. Four MS profiles were considered, corresponding to Clinical Isolated Syndrome (CIS), Relapsing-Remitting MS (RRMS), Secondary Progressive MS (SPMS), and Primary Progressive MS (PPMS). Each patient was classified in one of these profiles by our neurologist and underwent longitudinal MRI examinations including T1-weighted image acquisition at each examination, from which the GM tissue was segmented and the cortical GM thickness measured. Following the GM parcellation using two different atlases (FSAverage and Glasser 2016), the morphological connectome was built and six global metrics (Betweenness Centrality (BC), Assortativity (r), Transitivity (T), Efficiency (E g ), Modularity (Q) and Density (D)) were extracted. Based on their connectivity metrics, MS profiles were first statistically compared and second, classified using four different learning machines (Logistic Regression, Random Forest, Support Vector Machine and AdaBoost), combined in a higher level ensemble model by majority voting. Finally, the impact of the GM spatial resolution on the MS clinical profiles classification was analyzed. Results: Using binary comparisons between the four MS clinical profiles, statistical differences and classification performances higher than 0.7 were observed. Good performances were obtained when comparing the two early clinical forms, RRMS and PPMS (F1 score of 0.86), and the two neurodegenerative profiles, PPMS and SPMS (F1 score of 0.72). When comparing the two atlases, slightly better performances were obtained with the Glasser 2016 atlas, especially between RRMS with PPMS (F1 score of 0.83), compared to the FSAverage atlas (F1 score of 0.69). Also, the thresholding value for graph binarization was investigated suggesting more informative graph properties in the percentile range between 0.6 and 0.8. Conclusion: An automated pipeline was proposed for the classification of MS clinical profiles using six global graph metrics extracted from the GM morphological connectome of MS patients. This work demonstrated that GM morphological connectivity data could provide good classification performances by combining four simple ML models, without the cost of long and complex MR techniques, such as MR diffusion, and/or deep learning architectures.

Shape constrained CNN for segmentation guided prediction of myocardial shape and pose parameters in cardiac MRI.

Semantic segmentation using convolutional neural networks (CNNs) is the state-of-the-art for many medical image segmentation tasks including myocardial segmentation in cardiac MR images. However, the predicted segmentation maps obtained from such standard CNN do not allow direct quantification of regional shape properties such as regional wall thickness. Furthermore, the CNNs lack explicit shape constraints, occasionally resulting in unrealistic segmentations. In this paper, we use a CNN to predict shape parameters of an underlying statistical shape model of the myocardium learned from a training set of images. Additionally, the cardiac pose is predicted, which allows to reconstruct the myocardial contours. The integrated shape model regularizes the predicted contours and guarantees realistic shapes. We enforce robustness of shape and pose prediction by simultaneously performing pixel-wise semantic segmentation during training and define two loss functions to impose consistency between the two predicted representations: one distance-based loss and one overlap-based loss. We evaluated the proposed method in a 5-fold cross validation on an in-house clinical dataset with 75 subjects and on the ACDC and LVQuan19 public datasets. We show that the two newly defined loss functions successfully increase the consistency between shape and pose parameters and semantic segmentation, which leads to a significant improvement of the reconstructed myocardial contours. Additionally, these loss functions drastically reduce the occurrence of unrealistic shapes in the semantic segmentation output.

Corrigendum: Artificial Intelligence Based Patient-Specific Preoperative Planning Algorithm for Total Knee Arthroplasty.

[This corrects the article DOI: 10.3389/frobt.2022.840282.].

Prospective Natural History Study in 24 Adult Patients With LGMDR12 Over 2 Years of Follow-up: Quantitative MRI and Clinical Outcome Measures.

BACKGROUND AND OBJECTIVES: Limb-girdle muscular dystrophy autosomal recessive type 12 (LGMDR12) is a rare hereditary muscular dystrophy for which outcome measures are currently lacking. We evaluated quantitative MRI and clinical outcome measures to track disease progression to determine which tests could be useful in future clinical trials to evaluate potential therapies. METHODS: We prospectively measured the following outcome measures in all participants at baseline and after 1 and 2 years: 6-minute walk distance (6MWD), 10-meter walk test (10MWT), the Medical Research Council (MRC) sum scores, Biodex isometric dynamometry, serum creatine kinase, and 6-point Dixon MRI of the thighs. RESULTS: We included 24 genetically confirmed, adult patients with LGMDR12 and 24 age-matched and sex-matched healthy controls. Patients with intermediate-stage thigh muscle fat replacement at baseline (proton density fat fraction [PDFF] 20%-70%) already showed an increase in PDFF in 8 of the 14 evaluated thigh muscles after 1 year. The standardized response mean demonstrated a high responsiveness to change in PDFF for 6 individual muscles over 2 years in this group. However, in patients with early-stage (<20%) or end-stage (>70%) muscle fat replacement, PDFF did not increase significantly over 2 years of follow-up. Biodex isometric dynamometry showed a significant decrease in muscle strength in all patients in the right and left hamstrings (-6.2 Nm, p < 0.002 and -4.6 Nm, p < 0.009, respectively) and right quadriceps muscles (-9 Nm, p = 0.044) after 1 year of follow-up, whereas the 6MWD, 10MWT, and MRC sum scores were not able to detect a significant decrease in muscle function/strength even after 2 years. There was a moderately strong correlation between total thigh PDFF and clinical outcome measures at baseline. DISCUSSION: Thigh muscle PDFF imaging is a sensitive outcome measure to track progressive muscle fat replacement in selected patients with LGMDR12 even after 1 year of follow-up and correlates with clinical outcome measures. Biodex isometric dynamometry can reliably capture the loss of muscle strength over the course of 1 year in patients with LGMDR12 and should be included as an outcome measure in future clinical trials as well.

Treatment plan prediction for lung IMRT using deep learning based fluence map generation.

PURPOSE: Recently, it has been shown that automated treatment planning can be executed by direct fluence prediction from patient anatomy using convolutional neural networks. Proof of principle publications utilise a fixed dose prescription and fixed collimator (0°) and gantry angles. The goal of this work is to further develop these principles for the challenging lung cancer indication with variable dose prescriptions, collimator and gantry angles. First we investigate the impact of clinical applicable collimator angles and various input parameters. Then, the model is tested in a complete user independent planning workflow. METHODS: The dataset consists of 152 lung cancer patients, previously treated with IMRT. The patients are treated with either a left or a right beam setup and collimator angles and dose prescriptions adjusted to their tumour shape and stage. First we compare two CNNs with standard vs. personalised, clinical collimator angles. Next, four CNNs are trained with various combinations of CT and contour inputs. Finally, a complete user free treatment planning workflow is evaluated. RESULTS: The difference between the predicted and ground truth fluence maps for the fluence prediction CNN with all anatomical inputs in terms of the mean mean absolute error (MAE) is 4.17 × 10-4 for a fixed collimator angle and 5.46 × 10-4 for variable collimator angles. These differences vanish in terms of DVH metrics. Furthermore, the impact of anatomical inputs is small. The mean MAE is 5.88 × 10-4 if no anatomical information is given to the network. The DVH differences increase when a total user free planning workflow is examined. CONCLUSIONS: Fluence prediction with personalised collimator angles performs as good as fluence prediction with a standard collimator angle of zero degrees. The impact of anatomical inputs is small. The combination of a dose prediction and fluence prediction CNN deteriorates the fluence predictions. More investigation is required.

Artificial Intelligence Based Patient-Specific Preoperative Planning Algorithm for Total Knee Arthroplasty.

Previous studies have shown that the manufacturer's default preoperative plans for total knee arthroplasty with patient-specific guides require frequent, time-consuming changes by the surgeon. Currently, no research has been done on predicting preoperative plans for orthopedic surgery using machine learning. Therefore, this study aims to evaluate whether artificial intelligence (AI) driven planning tools can create surgeon and patient-specific preoperative plans that require fewer changes by the surgeon. A dataset of 5409 preoperative plans, including the manufacturer's default and the plans corrected by 39 surgeons, was collected. Features were extracted from the preoperative plans that describe the implant sizes, position, and orientation in a surgeon- and patient-specific manner. Based on these features, non-linear regression models were employed to predict the surgeon's corrected preoperative plan. The average number of corrections a surgeon has to make to the preoperative plan generated using AI was reduced by 39.7% compared to the manufacturer's default plan. The femoral and tibial implant size in the manufacturer's plan was correct in 68.4% and 73.1% of the cases, respectively, while the AI-based plan was correct in 82.2% and 85.0% of the cases, respectively, compared to the surgeon approved plan. Our method successfully demonstrated the use of machine learning to create preoperative plans in a surgeon- and patient-specific manner for total knee arthroplasty.

Cross-Modal Distillation to Improve MRI-Based Brain Tumor Segmentation With Missing MRI Sequences.

Convolutional neural networks (CNNs) for brain tumor segmentation are generally developed using complete sets of magnetic resonance imaging (MRI) sequences for both training and inference. As such, these algorithms are not trained for realistic, clinical scenarios where parts of the MRI sequences which were used for training, are missing during inference. To increase clinical applicability, we proposed a cross-modal distillation approach to leverage the availability of multi-sequence MRI data for training and generate an enriched CNN model which uses only single-sequence MRI data for inference but outperforms a single-sequence CNN model. We assessed the performance of the proposed method for whole tumor and tumor core segmentation with multi-sequence MRI data available for training but only T1-weighted ([Formula: see text]) sequence data available for inference, using BraTS 2018, and in-house datasets. Results showed that cross-modal distillation significantly improved the Dice score for both whole tumor and tumor core segmentation when only [Formula: see text] sequence data were available for inference. For the evaluation using the in-house dataset, cross-modal distillation achieved an average Dice score of 79.04% and 69.39% for whole tumor and tumor core segmentation, respectively, while a single-sequence U-Net model using [Formula: see text] sequence data for both training and inference achieved an average Dice score of 73.60% and 62.62%, respectively. These findings confirmed cross-modal distillation as an effective method to increase the potential of single-sequence CNN models such that segmentation performance is less compromised by missing MRI sequences or having only one MRI sequence available for segmentation.

Pitfalls in training and validation of deep learning systems.

The number of publications in endoscopic journals that present deep learning applications has risen tremendously over the past years. Deep learning has shown great promise for automated detection, diagnosis and quality improvement in endoscopy. However, the interdisciplinary nature of these works has undoubtedly made it more difficult to estimate their value and applicability. In this review, the pitfalls and common misconducts when training and validating deep learning systems are discussed and some practical guidelines are proposed that should be taken into account when acquiring data and handling it to ensure an unbiased system that will generalize for application in routine clinical practice. Finally, some considerations are presented to ensure correct validation and comparison of AI systems.

icobrain ms 5.1: Combining unsupervised and supervised approaches for improving the detection of multiple sclerosis lesions.

Multiple sclerosis (MS) is a chronic autoimmune, inflammatory neurological disease of the central nervous system. Its diagnosis nowadays commonly includes performing an MRI scan, as it is the most sensitive imaging test for MS. MS plaques are commonly identified from fluid-attenuated inversion recovery (FLAIR) images as hyperintense regions that are highly varying in terms of their shapes, sizes and locations, and are routinely classified in accordance to the McDonald criteria. Recent years have seen an increase in works that aimed at development of various semi-automatic and automatic methods for detection, segmentation and classification of MS plaques. In this paper, we present an automatic combined method, based on two pipelines: a traditional unsupervised machine learning technique and a deep-learning attention-gate 3D U-net network. The deep-learning network is specifically trained to address the weaker points of the traditional approach, namely difficulties in segmenting infratentorial and juxtacortical plaques in real-world clinical MRIs. It was trained and validated on a multi-center multi-scanner dataset that contains 159 cases, each with T1 weighted (T1w) and FLAIR images, as well as manual delineations of the MS plaques, segmented and validated by a panel of raters. The detection rate was quantified using lesion-wise Dice score. A simple label fusion is implemented to combine the output segmentations of the two pipelines. This combined method improves the detection of infratentorial and juxtacortical lesions by 14% and 31% respectively, in comparison to the unsupervised machine learning pipeline that was used as a performance assessment baseline.

68Ga-PSMA-11 PET, 18F-PSMA-1007 PET, and MRI for Gross Tumor Volume Delineation in Primary Prostate Cancer: Intermodality and Intertracer Variability.

PURPOSE: To assess the intermodality and intertracer variability of gallium-68 (68Ga)- or fluorine-18 (18F)-labeled prostate-specific membrane antigen (PSMA) positron emission tomography (PET) and biparametric magnetic resonance imaging (bpMRI)-based gross tumor volume (GTV) delineation for focal boosting in primary prostate cancer. METHODS: Nineteen prospectively enrolled patients with prostate cancer underwent a PSMA PET/MRI scan, divided into a 1:1 ratio between 68Ga-PSMA-11 and 18F-PSMA-1007, before radical prostatectomy (IWT140193). Four delineation teams performed manual contouring of the GTV based on bpMRI and PSMA PET imaging, separately. Index lesion coverage (overlap%) and interobserver variability were assessed. Furthermore, the distribution of the voxelwise normalized standardized uptake values (SUV%) was determined for the majority-voted (>50%) GTV (GTVmajority) and whole prostate gland to investigate intertracer variability. The median patientwise SUV% contrast ratio (SUV%-CR, calculated as median GTVmajority SUV% / median prostate gland without GTVmajority SUV%) was calculated according to the tracer used. RESULTS: A significant difference in overlap% favoring PSMA PET compared with bpMRI was found in the 18F subgroup (median, 63.0% vs 53.1%; P = .004) but was not present in the 68Ga subgroup (32.5% vs 50.6%; P = .100). Regarding interobserver variability, measured Sørensen-Dice coefficients (0.58 vs 0.72) and calculated mean distances to agreement (2.44 mm vs 1.22 mm) were statistically significantly lower and higher, respectively, for the 18F cohort compared with the 68Ga cohort. For the bpMRI-based delineations, the median Sørensen-Dice coefficient and mean distance to agreement were 0.63 and 1.76 mm, respectively. Median patientwise SUV%-CRs of 1.8 (interquartile range [IQR], 1.6-2.7) for 18F-PSMA and 3.3 (IQR, 2.7-5.9) for 68Ga-PSMA PET images were found. CONCLUSIONS: Both MRI and PSMA PET provided consistent intraprostatic GTV lesion detection. However, the PSMA tracer seems to have a major influence on the contour characteristics, owing to an apparent difference in SUV% distribution in the prostate gland.

Resorption of retromolar bone grafts after alveolar ridge augmentation-volumetric changes after 12 months assessed by CBCT analysis.

In this pilot study, a volumetric analysis of retromolar onlay bone grafts over a period of 12 months was conducted, using repeated CBCT imaging combined with automated image registration.Eleven patients being treated with 16 bone grafts taken from the retromolar area were examined by CBCT scanning prior to bone augmentation (T0), immediately after bone augmentation (T1) and after a healing time of 12 months after augmentation (T2). Graft volumes were measured at each time point after automated image registration of consecutive CBCT scans.The mean volume of the augmented site was 372.2 ± 179.4 mm3. Resorption relative to the original augmented volume was 43.7% ± 19.0% after 12 months.Three-dimensional graft resorption could be precisely depicted by the use of automated image registration for CBCT data over a period of 12 months and demonstrated extensive volumetric changes of bone grafts taken from the ascending ramus of the mandible.Graft resorption and continuous bony remodeling of the grafted site before and after implant insertion have to be carefully considered by the clinician.

Optimal 68Ga-PSMA and 18F-PSMA PET window levelling for gross tumour volume delineation in primary prostate cancer.

PURPOSE: This study proposes optimal tracer-specific threshold-based window levels for PSMA PET-based intraprostatic gross tumour volume (GTV) contouring to reduce interobserver delineation variability. METHODS: Nine 68Ga-PSMA-11 and nine 18F-PSMA-1007 PET scans including GTV delineations of four expert teams (GTVmanual) and a majority-voted GTV (GTVmajority) were assessed with respect to a registered histopathological GTV (GTVhisto) as the gold standard reference. The standard uptake values (SUVs) per voxel were converted to a percentage (SUV%) relative to the SUVmax. The statistically optimised SUV% threshold (SOST) was defined as those that maximises accuracy for threshold-based contouring. A leave-one-out cross-validation receiver operating characteristic (ROC) curve analysis was performed to determine the SOST for each tracer. The SOST analysis was performed twice, first using the GTVhisto contour as training structure (GTVSOST-H) and second using the GTVmajority contour as training structure (GTVSOST-MA) to correct for any limited misregistration. The accuracy of both GTVSOST-H and GTVSOST-MA was calculated relative to GTVhisto in the 'leave-one-out' patient of each fold and compared with the accuracy of GTVmanual. RESULTS: ROC curve analysis for 68Ga-PSMA-11 PET revealed a median threshold of 25 SUV% (range, 22-27 SUV%) and 41 SUV% (40-43 SUV%) for GTVSOST-H and GTVSOST-MA, respectively. For 18F-PSMA-1007 PET, a median threshold of 42 SUV% (39-45 SUV%) for GTVSOST-H and 44 SUV% (42-45 SUV%) for GTVSOST-MA was found. A significant pairwise difference was observed when comparing the accuracy of the GTVSOST-H contours with the median accuracy of the GTVmanual contours (median, - 2.5%; IQR, - 26.5-0.2%; p = 0.020), whereas no significant pairwise difference was found for the GTVSOST-MA contours (median, - 0.3%; IQR, - 4.4-0.6%; p = 0.199). CONCLUSIONS: Threshold-based contouring using GTVmajority-trained SOSTs achieves an accuracy comparable with manual contours in delineating GTVhisto. The median SOSTs of 41 SUV% for 68Ga-PSMA-11 PET and 44 SUV% for 18F-PSMA-1007 PET form a base for tracer-specific window levelling. TRIAL REGISTRATION: Clinicaltrials.gov ; NCT03327675; 31-10-2017.