Dice Similarity Coefficient Scores Research Articles

Multimodal Data-Driven Segmentation of Bone Metastasis Lesions in SPECT Bone Scans Using Deep Learning

Background: Patients with malignant tumors often develop bone metastases. SPECT bone scintigraphy is an effective tool for surveying bone metastases due to its high sensitivity, low-cost equipment, and radiopharmaceutical. However, the low spatial resolution of SPECT scans significantly hinders manual analysis by nuclear medicine physicians. Deep learning, a promising technique for automated image analysis, can extract hierarchal patterns from images without human intervention. Objective: To enhance the performance of deep learning-based segmentation models, we integrate textual data from diagnostic reports with SPECT bone scans, aiming to develop an automated analysis method that outperforms purely unimodal data-driven segmentation models. Methods: We propose a dual-path segmentation framework to extract features from bone scan images and diagnostic reports separately. In the first path, an encoder-decoder network is employed to learn hierarchical representations of features from SPECT bone scan images. In the second path, the Chinese version of the MacBERT model is utilized to develop a text encoder for extracting features from diagnostic reports. The extracted textual features are then fused with image features during the decoding stage in the first path, enhancing the overall segmentation performance. Results: Experimental evaluation conducted on real-world clinical data demonstrated the superior performance of the proposed segmentation model. Our model achieved a 0.0209 increase in the DSC (Dice Similarity Coefficient) score compared to the well-known U-Net model. Conclusions: The proposed multimodal data-driven method effectively identifies and isolates metastasis lesions in SPECT bone scans, outperforming existing classical deep learning models. This study demonstrates the value of incorporating textual data in the deep learning-based segmentation of lowresolution SPECT bone scans.

Precise ablation zone segmentation on CT images after liver cancer ablation using semi‐automatic CNN‐based segmentation

AbstractBackgroundAblation zone segmentation in contrast‐enhanced computed tomography (CECT) images enables the quantitative assessment of treatment success in the ablation of liver lesions. However, fully automatic liver ablation zone segmentation in CT images still remains challenging, such as low accuracy and time‐consuming manual refinement of the incorrect regions.PurposeTherefore, in this study, we developed a semi‐automatic technique to address the remaining drawbacks and improve the accuracy of the liver ablation zone segmentation in the CT images.MethodsOur approach uses a combination of a CNN‐based automatic segmentation method and an interactive CNN‐based segmentation method. First, automatic segmentation is applied for coarse ablation zone segmentation in the whole CT image. Human experts then visually validate the segmentation results. If there are errors in the coarse segmentation, local corrections can be performed on each slice via an interactive CNN‐based segmentation method. The models were trained and the proposed method was evaluated using two internal datasets of post‐interventional CECT images ( = 22, = 145; 62 patients in total) and then further tested using an external benchmark dataset ( = 12; 10 patients).ResultsTo evaluate the accuracy of the proposed approach, we used Dice similarity coefficient (DSC), average symmetric surface distance (ASSD), Hausdorff distance (HD), and volume difference (VD). The quantitative evaluation results show that the proposed approach obtained mean DSC, ASSD, HD, and VD scores of 94.0%, 0.4 mm, 8.4 mm, 0.02, respectively, on the internal dataset, and 87.8%, 0.9 mm, 9.5 mm, and −0.03, respectively, on the benchmark dataset. We also compared the performance of the proposed approach to that of five well‐known segmentation methods; the proposed semi‐automatic method achieved state‐of‐the‐art performance on ablation segmentation accuracy, and on average, 2 min are required to correct the segmentation. Furthermore, we found that the accuracy of the proposed method on the benchmark dataset is comparable to that of manual segmentation by human experts ( = 0.55, ‐test).ConclusionsThe proposed semi‐automatic CNN‐based segmentation method can be used to effectively segment the ablation zones, increasing the value of CECT for an assessment of treatment success. For reproducibility, the trained models, source code, and demonstration tool are publicly available at https://github.com/lqanh11/Interactive_AblationZone_Segmentation.

Accuracy of deep learning-based upper airway segmentation

IntroductionIn orthodontic treatments, accurately assessing the upper airway volume and morphology is essential for proper diagnosis and planning. Cone beam computed tomography (CBCT) is used for assessing upper airway volume through manual, semi-automatic, and automatic airway segmentation methods. This study evaluates upper airway segmentation accuracy by comparing the results of an automatic model and a semi-automatic method against the gold standard manual method. Materials and methodsAn automatic segmentation model was trained using the MONAI Label framework to segment the upper airway from CBCT images. An open-source program, ITK-SNAP, was used for semi-automatic segmentation. The accuracy of both methods was evaluated against manual segmentations. Evaluation metrics included Dice Similarity Coefficient (DSC), Precision, Recall, 95% Hausdorff Distance (HD), and volumetric differences. ResultsThe automatic segmentation group averaged a DSC score of 0.915±0.041, while the semi-automatic group scored 0.940±0.021, indicating clinically acceptable accuracy for both methods. Analysis of the 95% HD revealed that semi-automatic segmentation (0.997±0.585) was more accurate and closer to manual segmentation than automatic segmentation (1.447±0.674). Volumetric comparisons revealed no statistically significant differences between automatic and manual segmentation for total, oropharyngeal, and velopharyngeal airway volumes. Similarly, no significant differences were noted between the semi-automatic and manual methods across these regions. ConclusionIt has been observed that both automatic and semi-automatic methods, which utilise open-source software, align effectively with manual segmentation. Implementing these methods can aid in decision-making by allowing faster and easier upper airway segmentation with comparable accuracy in orthodontic practice.

Head and neck automatic multi-organ segmentation on Dual-Energy Computed Tomography

Background and purposeDeep-learning-based automatic segmentation is widely used in radiation oncology to delineate organs-at-risk. Dual-energy CT (DECT) allows the reconstruction of enhanced contrast images that could help with manual and auto-delineation. This paper presents a performance evaluation of a commercial auto-segmentation software on image series generated by a DECT. Material and methodsDifferent types of DECT images from 74 head-and-neck (HN) patients were retrieved, including polyenergetic images at different voltages [80 kV with DirectDensity (PEI80-DD), 80 kV (PEI80), 120 kV-mixed (PEI120)] and a virtual-monoenergetic image at 40 keV (VMI40). Delineations used for treatment planning were considered as ground truth (GT) and were compared with the auto-segmentations performed on the 4 DECT images. A blinded qualitative evaluation of 3 structures (thyroid, left parotid, left nodes level II) was carried out. Performance metrics were calculated for 13 HN structures to evaluate the auto-contours including dice similarity coefficient (DSC), 95th percentile Hausdorff distance (95HD) and mean surface distance (MSD). ResultsWe observed a high rate of low scores for PEI80-DD and VMI40 auto-segmentations on the thyroid and for GT and VMI40 contours on the nodes level II. All images received excellent scores for the parotid glands. The metrics comparison between GT and auto-segmented contours revealed that PEI80-DD had the highest DSC scores, significantly outperforming other reconstructed images for all organs (p < 0.05). ConclusionsThe results indicate that the auto-contouring system cannot generalize to images derived from DECT acquisition. It is therefore crucial to identify which organs benefit from these acquisitions to adapt the training datasets accordingly.

Impact of the Novel MRI Contrast Agent Gadopiclenol on Radiotherapy Decision Making in Patients With Brain Metastases.

The aim of this study was to assess the effect of gadopiclenol versus gadobenate dimeglumine contrast-enhanced magnetic resonance imaging (MRI) on decision-making between whole-brain radiotherapy (WBRT) and stereotactic radiosurgery (SRS) for treatment of brain metastases (BMs). Patients with BMs underwent 2 separate MRI examinations in a double-blind crossover phase IIb comparative study between the MRI contrast agents gadopiclenol and gadobenate dimeglumine, both administered at 0.1 mmol/kg. The imaging data of a single site using identical MRI scanners and protocols were included in this post hoc analysis. Patients with 1 or more BMs in any of both MRIs were subjected to target volume delineation for treatment planning. Two radiation oncologists contoured all visible lesions and decided upon SRS or WBRT, according to the number of metastases. For each patient, SRS or WBRT treatment plans were calculated for both MRIs, considering the gross target volume (GTV) as the contrast-enhancing aspects of the tumor. Mean GTVs and volume of healthy brain exposed to 12 Gy (V12), as well as Dice similarity coefficient scores, were obtained. The Spearman rank (ρ) correlation was additionally calculated for assessing linear differences. Three different expert radiation oncologists blindly rated the contrast enhancement for contouring purposes. Thirteen adult patients were included. Gadopiclenol depicted additional BM as compared with gadobenate dimeglumine in 7 patients (54%). Of a total of 63 identified metastatic lesions in both MRI sets, 3 subgroups could be defined: A, 48 (24 pairs) detected equal GTVs visible in both modalities; B, 13 GTVs only visible in the gadopiclenol set (mean ± SD, 0.16 ± 0.37 cm3); and C, 2 GTVs only visible in the gadobenate dimeglumine set (mean ± SD, 0.01 ± 0.01). Treatment indication was changed for 2 (15%) patients, 1 from no treatment to SRS and for 1 from SRS to WBRT. The mean GTVs and brain V12 were comparable between both agents (P = 0.694, P = 0.974). The mean Dice similarity coefficient was 0.70 ± 0.14 (ρ = 0.82). According to the readers, target volume definition was improved in 63.9% of cases (23 of 36 evaluations) with gadopiclenol and 22.2% with gadobenate dimeglumine (8 of 36), whereas equivalence was obtained in 13.9% (5 of 36). Gadopiclenol-enhanced MRI improved BM detection and characterization, with a direct impact on radiotherapy treatment decision between WBRT and SRS. Additionally, a more exact target delineation and planning could be performed with gadopiclenol. A prospective evaluation in a larger cohort of patients is required to confirm these findings.

A multi-task generative model for simultaneous post-contrast MR image synthesis and brainstem glioma segmentation

ObjectivesThis study aims to generate post-contrast MR images reducing the exposure of gadolinium-based contrast agents (GBCAs) for brainstem glioma (BSG) detection, simultaneously delineating the BSG lesion, and providing high-resolution contrast information. MethodsA retrospective cohort of 30 patients diagnosed with brainstem glioma was included. Multi-contrast images, including pre-contrast T1 weighted (pre-T1w), T2 weighted (T2w), arterial spin labeling (ASL) and post-contrast T1w images, were collected. A multi-task generative model was developed to synthesize post-contrast T1w images and simultaneously segment BSG masks from the multi-contrast inputs. Performance evaluation was conducted using peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean absolute error (MAE) metrics. A perceptual study was also undertaken to assess diagnostic quality. ResultsThe proposed model achieved SSIM of 0.86 ± 0.04, PSNR of 26.33 ± 0.05 and MAE of 57.20 ± 20.50 for post-contrast T1w image synthesis. Automated delineation of the BSG lesions achieved Dice similarity coefficient (DSC) score of 0.88 ± 0.27. ConclusionsThe proposed model can synthesize high-quality post-contrast T1w images and accurately segment the BSG region, yielding satisfactory DSC scores. Clinical relevance statementThe synthesized post-contrast MR image presented in this study has the potential to reduce the usage of gadolinium-based contrast agents, which may pose risks to patients. Moreover, the automated segmentation method proposed in this paper aids radiologists in accurately identifying the brainstem glioma lesion, facilitating the diagnostic process.

Towards reliable healthcare Imaging: conditional contrastive generative adversarial network for handling class imbalancing in MR Images.

Medical imaging datasets frequently encounter a data imbalance issue, where the majority of pixels correspond to healthy regions, and the minority belong to affected regions. This uneven distribution of pixels exacerbates the challenges associated with computer-aided diagnosis. The networks trained with imbalanced data tends to exhibit bias toward majority classes, often demonstrate high precision but low sensitivity. We have designed a new network based on adversarial learning namely conditional contrastive generative adversarial network (CCGAN) to tackle the problem of class imbalancing in a highly imbalancing MRI dataset. The proposed model has three new components: (1) class-specific attention, (2) region rebalancing module (RRM) and supervised contrastive-based learning network (SCoLN). The class-specific attention focuses on more discriminative areas of the input representation, capturing more relevant features. The RRM promotes a more balanced distribution of features across various regions of the input representation, ensuring a more equitable segmentation process. The generator of the CCGAN learns pixel-level segmentation by receiving feedback from the SCoLN based on the true negative and true positive maps. This process ensures that final semantic segmentation not only addresses imbalanced data issues but also enhances classification accuracy. The proposed model has shown state-of-art-performance on five highly imbalance medical image segmentation datasets. Therefore, the suggested model holds significant potential for application in medical diagnosis, in cases characterized by highly imbalanced data distributions. The CCGAN achieved the highest scores in terms of dice similarity coefficient (DSC) on various datasets: 0.965±0.012 for BUS2017, 0.896±0.091 for DDTI, 0.786±0.046 for LiTS MICCAI 2017, 0.712±1.5 for the ATLAS dataset, and 0.877±1.2 for the BRATS 2015 dataset. DeepLab-V3 follows closely, securing the second-best position with DSC scores of 0.948±0.010 for BUS2017, 0.895±0.014 for DDTI, 0.763±0.044 for LiTS MICCAI 2017, 0.696±1.1 for the ATLAS dataset, and 0.846±1.4 for the BRATS 2015 dataset.

CapNet: An Automatic Attention-Based with Mixer Model for Cardiovascular Magnetic Resonance Image Segmentation.

Deep neural networks have shown excellent performance in medical image segmentation, especially for cardiac images. Transformer-based models, though having advantages over convolutional neural networks due to the ability of long-range dependence learning, still have shortcomings such as having a large number of parameters and and high computational cost. Additionally, for better results, they are often pretrained on a larger data, thus requiring large memory size and increasing resource expenses. In this study, we propose a new lightweight but efficient model, namely CapNet, based on convolutions and mixing modules for cardiac segmentation from magnetic resonance images (MRI) that can be trained from scratch with a small amount of parameters. To handle varying sizes and shapes which often occur in cardiac systolic and diastolic phases, we propose attention modules for pooling, spatial, and channel information. We also propose a novel loss called the Tversky Shape Power Distance function based on the shape dissimilarity between labels and predictions that shows promising performances compared to other losses. Experiments on three public datasets including ACDC benchmark, Sunnybrook data, and MS-CMR challenge are conducted and compared with other state of the arts (SOTA). For binary segmentation, the proposed CapNet obtained the Dice similarity coefficient (DSC) of 94% and 95.93% for respectively the Endocardium and Epicardium regions with Sunnybrook dataset, 94.49% for Endocardium, and 96.82% for Epicardium with the ACDC data. Regarding the multiclass case, the average DSC by CapNet is 93.05% for the ACDC data; and the DSC scores for the MS-CMR are 94.59%, 92.22%, and 93.99% for respectively the bSSFP, T2-SPAIR, and LGE sequences of the MS-CMR. Moreover, the statistical significance analysis tests with p-value compared with transformer-based methods and some CNN-based approaches demonstrated that the CapNet, though having fewer training parameters, is statistically significant. The promising evaluation metrics show comparative results in both Dice and IoU indices compared to SOTA CNN-based and Transformer-based architectures.

Automatic 3D left atrial strain extraction framework on cardiac computed tomography

Background and objectiveStrain analysis provides insights into myocardial function and cardiac condition evaluation. However, the anatomical characteristics of left atrium (LA) inherently limit LA strain analysis when using echocardiography. Cardiac computed tomography (CT) with its superior spatial resolution, has become critical for in-depth evaluation of LA function. Recent studies have explored the feasibility of CT-derived strain; however, they relied on manually selected regions of interest (ROIs) and mainly focused on left ventricle (LV). This study aimed to propose a first-of-its-kind fully automatic deep learning (DL)-based framework for three-dimensional (3D) LA strain extraction on cardiac CT. MethodsA total of 111 patients undergoing ECG-gated contrast-enhanced CT for evaluating subclinical atrial fibrillation (AF) were enrolled in this study. We developed a 3D strain extraction framework on cardiac CT images, containing a 2.5D GN-U-Net network for LA segmentation, axis-oriented 3D view extraction, and LA strain measure. The segmentation accuracy was evaluated using Dice similarity coefficient (DSC). The model-extracted LA volumes and emptying fraction (EF) were compared with ground-truth measurements using intraclass correlation coefficient (ICC), correlation coefficient (r), and Bland-Altman plot (B-A). The automatically extracted LA strains were evaluated against the LA strains measured from 2D echocardiograms. We utilized this framework to gauge the effect of AF burden on LA strain, employing the atrial high rate episode (AHRE) burden as the measurement parameter. ResultsThe GN-U-Net LA segmentation network achieved a DSC score of 0.9603 on the test set. The framework-extracted LA estimates demonstrated excellent ICCs of 0.949 (95 % CI: 0.93–0.97) for minimal LA volume, 0.904 (95 % CI: 0.86–0.93) for maximal LA volume, and 0.902 (95 % CI: 0.86–0.93) for EF, compared with expert measurements. The framework-extracted LA strains demonstrated moderate agreement with the LA strains based on 2D echocardiography (ICCs >0.703). Patients with AHRE > 6 min had significantly lower global strain and LAEF, as extracted by the framework than those with AHRE ≤ 6 min. ConclusionThe promising results highlighted the feasibility and clinical usefulness of automatically extracting 3D LA strain from CT images using a DL-based framework. This tool could provide a 3D-based alternative to echocardiography for assessing LA function.

Automated prostate gland segmentation in challenging clinical cases: comparison of three artificial intelligence methods.

Automated methods for prostate segmentation on MRI are typically developed under ideal scanning and anatomical conditions. This study evaluates three different prostate segmentation AI algorithms in a challenging population of patients with prior treatments, variable anatomic characteristics, complex clinical history, or atypical MRI acquisition parameters. A single institution retrospective database was queried for the following conditions at prostate MRI: prior prostate-specific oncologic treatment, transurethral resection of the prostate (TURP), abdominal perineal resection (APR), hip prosthesis (HP), diversity of prostate volumes (large ≥ 150cc, small ≤ 25cc), whole gland tumor burden, magnet strength, noted poor quality, and various scanners (outside/vendors). Final inclusion criteria required availability of axial T2-weighted (T2W) sequence and corresponding prostate organ segmentation from an expert radiologist. Three previously developed algorithms were evaluated: (1) deep learning (DL)-based model, (2) commercially available shape-based model, and (3) federated DL-based model. Dice Similarity Coefficient (DSC) was calculated compared to expert. DSC by model and scan factors were evaluated with Wilcox signed-rank test and linear mixed effects (LMER) model. 683 scans (651 patients) met inclusion criteria (mean prostate volume 60.1cc [9.05-329cc]). Overall DSC scores for models 1, 2, and 3 were 0.916 (0.707-0.971), 0.873 (0-0.997), and 0.894 (0.025-0.961), respectively, with DL-based models demonstrating significantly higher performance (p < 0.01). In sub-group analysis by factors, Model 1 outperformed Model 2 (all p < 0.05) and Model 3 (all p < 0.001). Performance of all models was negatively impacted by prostate volume and poor signal quality (p < 0.01). Shape-based factors influenced DL models (p < 0.001) while signal factors influenced all (p < 0.001). Factors affecting anatomical and signal conditions of the prostate gland can adversely impact both DL and non-deep learning-based segmentation models.

CGMA-Net: Cross-Level Guidance and Multi-Scale Aggregation Network for Polyp Segmentation.

Colonoscopy is considered the best prevention and control method for colorectal cancer, which suffers extremely high rates of mortality and morbidity. Automated polyp segmentation of colonoscopy images is of great importance since manual polyp segmentation requires a considerable time of experienced specialists. However, due to the high similarity between polyps and mucosa, accompanied by the complex morphological features of colonic polyps, the performance of automatic polyp segmentation is still unsatisfactory. Accordingly, we propose a network, namely Cross-level Guidance and Multi-scale Aggregation (CGMA-Net), to earn a performance promotion. Specifically, three modules, including Cross-level Feature Guidance (CFG), Multi-scale Aggregation Decoder (MAD), and Details Refinement (DR), are individually proposed and synergistically assembled. With CFG, we generate spatial attention maps from the higher-level features and then multiply them with the lower-level features, highlighting the region of interest and suppressing the background information. In MAD, we parallelly use multiple dilated convolutions of different sizes to capture long-range dependencies between features. For DR, an asynchronous convolution is used along with the attention mechanism to enhance both the local details and the global information. The proposed CGMA-Net is evaluated on two benchmark datasets, i.e., CVC-ClinicDB and Kvasir-SEG, whose results demonstrate that our method not only presents state-of-the-art performance but also holds relatively fewer parameters. Concretely, we achieve the Dice Similarity Coefficient (DSC) of 91.85% and 95.73% on Kvasir-SEG and CVC-ClinicDB, respectively. The assessment of model generalization is also conducted, resulting in DSC scores of 86.25% and 86.97% on the two datasets respectively.

The role of 18F-FDG PET in minimizing variability in gross tumor volume delineation of soft tissue sarcomas

BackgroundAccurate gross tumor volume (GTV) delineation is a critical step in radiation therapy treatment planning. However, it is reader dependent and thus susceptible to intra- and inter-reader variability. GTV delineation of soft tissue sarcoma (STS) often relies on CT and MR images. PurposeThis study investigates the potential role of 18F-FDG PET in reducing intra- and inter-reader variability thereby improving reproducibility of GTV delineation in STS, without incurring additional costs or radiation exposure. Materials and MethodsThree readers performed independent GTV delineation of 61 patients with STS using first CT and MR followed by CT, MR, and 18F-FDG PET images. Each reader performed a total of six delineation trials, three trials per imaging modality group. Dice Similarity Coefficient (DSC) score and Hausdorff distance (HD) were used to assess both intra- and inter-reader variability using generated simultaneous truth and performance level estimation (STAPLE) GTVs as ground truth. Statistical analysis was performed using a Wilcoxon signed-ranked test. ResultsThere was a statistically significant decrease in both intra- and inter-reader variability in GTV delineation using CT, MR 18F-FDG PET images vs. CT and MR images. This was translated by an increase in the DSC score and a decrease in the HD for GTVs drawn from CT, MR and 18F-FDG PET images vs. GTVs drawn from CT and MR for all readers and across all three trials. ConclusionIncorporation of 18F-FDG PET into CT and MR images decreased intra- and inter-reader variability and subsequently increased reproducibility of GTV delineation in STS.

Automatic segmentation of skeletal muscles from MR images using modified U-Net and a novel data augmentation approach.

Rapid and accurate muscle segmentation is essential for the diagnosis and monitoring of many musculoskeletal diseases. As gold standard, manual annotation suffers from intensive labor and high inter-operator reproducibility errors. In this study, deep learning (DL) based automatic muscle segmentation from MR scans is investigated for post-menopausal women, who normally experience a decline in muscle volume. The performance of four Deep Learning (DL) models was evaluated: U-Net and UNet++ and two modified U-Net networks, which combined feature fusion and attention mechanisms (Feature-Fusion-UNet, FFU, and Attention-Feature-Fusion-UNet, AFFU). The models were tested for automatic segmentation of 16-lower limb muscles from MRI scans of two cohorts of post-menopausal women (11 subjects in PMW-1, 8 subjects in PMW-2; from two different studies so considered independent datasets) and 10 obese post-menopausal women (PMW-OB). Furthermore, a novel data augmentation approach is proposed to enlarge the training dataset. The results were assessed and compared by using the Dice similarity coefficient (DSC), relative volume error (RVE), and Hausdorff distance (HD). The best performance among all four DL models was achieved by AFFU (PMW-1: DSC 0.828 ± 0.079, 1-RVE 0.859 ± 0.122, HD 29.9mm ± 26.5mm; PMW-2: DSC 0.833 ± 0.065, 1-RVE 0.873 ± 0.105, HD 25.9mm ± 27.9mm; PMW-OB: DSC 0.862 ± 0.048, 1-RVE 0.919 ± 0.076, HD 34.8mm ± 46.8mm). Furthermore, the augmentation of data significantly improved the DSC scores of U-Net and AFFU for all 16 tested muscles (between 0.23% and 2.17% (DSC), 1.6%-1.93% (1-RVE), and 9.6%-19.8% (HD) improvement). These findings highlight the feasibility of utilizing DL models for automatic segmentation of muscles in post-menopausal women and indicate that the proposed augmentation method can enhance the performance of models trained on small datasets.

Advancements in cardiac structures segmentation: a comprehensive systematic review of deep learning in CT imaging.

Segmentation of cardiac structures is an important step in evaluation of the heart on imaging. There has been growing interest in how artificial intelligence (AI) methods-particularly deep learning (DL)-can be used to automate this process. Existing AI approaches to cardiac segmentation have mostly focused on cardiac MRI. This systematic review aimed to appraise the performance and quality of supervised DL tools for the segmentation of cardiac structures on CT. Embase and Medline databases were searched to identify related studies from January 1, 2013 to December 4, 2023. Original research studies published in peer-reviewed journals after January 1, 2013 were eligible for inclusion if they presented supervised DL-based tools for the segmentation of cardiac structures and non-coronary great vessels on CT. The data extracted from eligible studies included information about cardiac structure(s) being segmented, study location, DL architectures and reported performance metrics such as the Dice similarity coefficient (DSC). The quality of the included studies was assessed using the Checklist for Artificial Intelligence in Medical Imaging (CLAIM). 18 studies published after 2020 were included. The DSC scores median achieved for the most commonly segmented structures were left atrium (0.88, IQR 0.83-0.91), left ventricle (0.91, IQR 0.89-0.94), left ventricle myocardium (0.83, IQR 0.82-0.92), right atrium (0.88, IQR 0.83-0.90), right ventricle (0.91, IQR 0.85-0.92), and pulmonary artery (0.92, IQR 0.87-0.93). Compliance of studies with CLAIM was variable. In particular, only 58% of studies showed compliance with dataset description criteria and most of the studies did not test or validate their models on external data (81%). Supervised DL has been applied to the segmentation of various cardiac structures on CT. Most showed similar performance as measured by DSC values. Existing studies have been limited by the size and nature of the training datasets, inconsistent descriptions of ground truth annotations and lack of testing in external data or clinical settings. [www.crd.york.ac.uk/prospero/], PROSPERO [CRD42023431113].

Deep learning application for abdominal organs segmentation on 0.35 T MR-Linac images.

Linear accelerator (linac) incorporating a magnetic resonance (MR) imaging device providing enhanced soft tissue contrast is particularly suited for abdominal radiation therapy. In particular, accurate segmentation for abdominal tumors and organs at risk (OARs) required for the treatment planning is becoming possible. Currently, this segmentation is performed manually by radiation oncologists. This process is very time consuming and subject to inter and intra operator variabilities. In this work, deep learning based automatic segmentation solutions were investigated for abdominal OARs on 0.35 T MR-images. One hundred and twenty one sets of abdominal MR images and their corresponding ground truth segmentations were collected and used for this work. The OARs of interest included the liver, the kidneys, the spinal cord, the stomach and the duodenum. Several UNet based models have been trained in 2D (the Classical UNet, the ResAttention UNet, the EfficientNet UNet, and the nnUNet). The best model was then trained with a 3D strategy in order to investigate possible improvements. Geometrical metrics such as Dice Similarity Coefficient (DSC), Intersection over Union (IoU), Hausdorff Distance (HD) and analysis of the calculated volumes (thanks to Bland-Altman plot) were performed to evaluate the results. The nnUNet trained in 3D mode achieved the best performance, with DSC scores for the liver, the kidneys, the spinal cord, the stomach, and the duodenum of 0.96 ± 0.01, 0.91 ± 0.02, 0.91 ± 0.01, 0.83 ± 0.10, and 0.69 ± 0.15, respectively. The matching IoU scores were 0.92 ± 0.01, 0.84 ± 0.04, 0.84 ± 0.02, 0.54 ± 0.16 and 0.72 ± 0.13. The corresponding HD scores were 13.0 ± 6.0mm, 16.0 ± 6.6mm, 3.3 ± 0.7mm, 35.0 ± 33.0mm, and 42.0 ± 24.0mm. The analysis of the calculated volumes followed the same behavior. Although the segmentation results for the duodenum were not optimal, these findings imply a potential clinical application of the 3D nnUNet model for the segmentation of abdominal OARs for images from 0.35 T MR-Linac.

Differential Diagnosis of Benign and Malignant Pulmonary Nodules in CT Images Based on Multitask Learning.

Artificial intelligence-based aided diagnostic systems for pulmonary nodules can be divided into subtasks such as nodule detection, segmentation, and benign and malignant differentiation. Most current studies are limited to single-target tasks. However, aided diagnosis aims to distinguish benign from malignant pulmonary nodules, which requires the fusion of multiple-scale features and comprehensive discrimination based on the results of multiple learning tasks. This study focuses on the aspects of model design, network structure, and constraints and proposes a novel model that integrates the learning tasks of pulmonary nodule detection, segmentation, and classification under weakly supervised conditions. The main innovations include the following three aspects: (1) a two-dimensional sequence detection model based on a ConvLSTM (Convolutional Long Short-Term Memory) network and U-shaped structure network is proposed to obtain the context space features of image slices fully; (2) a differential diagnosis of benign and malignant pulmonary nodules based on multitask learning is proposed, which uses the annotated data of different types of tasks to mine the potential common features among tasks; and (3) an optimization strategy incorporating prior knowledge of computed tomography images and dynamic weight adjustment of multiple tasks is proposed to ensure that each task can efficiently complete training and learning. Experiments on the LIDC-IDRI and LUNA16 datasets showed that our proposed method achieved a final competition performance metric score of 87.80% for nodule detection and a Dice similarity coefficient score of 83.95% for pulmonary nodule segmentation. The cross-validation results of the LIDC-IDRI and LUNA16 datasets show that our model achieved 87.80% of the final competition performance metric score for nodule detection and 83.95% of the DSC score for pulmonary nodule segmentation, representing the optimal result for that dataset.

Evaluating the clinical applicability of neural networks for meningioma tumor segmentation on 3D MRI

Magnetic resonance imaging (MRI), in both 2D and 3D configurations, serves as the leading non-invasive and non-ionizing technique for the detection and clinical assessment of brain tumors. 3D multiparametric MRIs (mpMRIs) offer enhanced spatial and biological context over other MRIs, which can help clinicians plan tailored treatment and therapy. Presently, surgical intervention remains the primary treatment method for brain tumors, requiring accurate segmentation of tumor tissue on MRI scans. However, manual tumor segmentation and traditional machine learning techniques are labor intensive and subject to observer bias. Deep learning models can provide enhanced accuracy and precision in tumor segmentation due to their ability to automatically extract intricate features and patterns from MRI scans. In this study, MRI scans from 358 meningioma patients were used to train a series of deep learning models to segment the tumor subregions. We hypothesized that a deep learning framework designed for meningioma tumor segmentation and trained utilizing 3D mpMRIs would demonstrate superior segmentation accuracy compared to a framework trained on 2D MRIs or single-sequence 3D MRIs. We compared the model predictions with the ground truth labels using similarity metrics to assess accuracy and clinical applicability. The model trained on 3D mpMRI scans showed reliable performance for meningioma subregion segmentation with a Dice Similarity Coefficient score of 0.91 and a median Sensitivity of 91.38%. It outperformed the other models due to the additional spatial context provided by 3D mpMRIs, suggesting it could be readily used in clinical practice to help guide treatment strategy.

Artificial Intelligence–Based Autosegmentation: Advantages in Delineation, Absorbed Dose-Distribution, and Logistics

PurposeTo compare the performance of artificial intelligence (AI) in auto-contouring compared to a human practitioner in terms of precision, differences in dose distribution and time consumption. Methods and MaterialsDatasets of previously irradiated patients in three different segments (head and neck- (HNC), breast- (BC) and prostate cancer (PC)) were retrospectively collected. An experienced radiation oncologist (MD) performed organs-at-risk (OARs) and standard clinical target volume (CTV) delineations as baseline structures for comparison. AI-based auto-contours were generated in two additional CT copies; therefore, three groups were assessed: MD alone, AI alone, AI plus MD corrections (AI+C). Differences in Dice similarity coefficient (DSC) and person-hour burden were assessed. Furthermore, changes in clinically relevant dose-volume parameters were evaluated and compared. ResultsSeventy-five previously treated cases were collected (25 per segment) for the analysis. When compared to MD contours, the mean DSC scores were higher than 0.7 for 74% and 80% of AI and AI+C, respectively. After corrections, 17.1% structures presented DSC score deviations higher than 0.1 and 10.4% dose-volume parameters significantly changed in AI-contoured structures. The time consumption assessment yielded mean person-hour reductions of 68%, 51% and 71% for BC, PC and HNC, respectively. ConclusionIn great extent, AI yielded clinically acceptable OARs and certain CTVs in the explored anatomical segments. Sparse correction and assessment requirements place AI+C as a standard workflow. Minimal clinically relevant differences in OAR-exposure were identified. A substantial amount of person-hours could be repurposed with this technology.

MUE-CoT: multi-scale uncertainty entropy-aware co-training framework for left atrial segmentation

Objective. Accurate left atrial segmentation is the basis of the recognition and clinical analysis of atrial fibrillation. Supervised learning has achieved some competitive segmentation results, but the high annotation cost often limits its performance. Semi-supervised learning is implemented from limited labeled data and a large amount of unlabeled data and shows good potential in solving practical medical problems. Approach. In this study, we proposed a collaborative training framework for multi-scale uncertain entropy perception (MUE-CoT) and achieved efficient left atrial segmentation from a small amount of labeled data. Based on the pyramid feature network, learning is implemented from unlabeled data by minimizing the pyramid prediction difference. In addition, novel loss constraints are proposed for co-training in the study. The diversity loss is defined as a soft constraint so as to accelerate the convergence and a novel multi-scale uncertainty entropy calculation method and a consistency regularization term are proposed to measure the consistency between prediction results. The quality of pseudo-labels cannot be guaranteed in the pre-training period, so a confidence-dependent empirical Gaussian function is proposed to weight the pseudo-supervised loss. Main results. The experimental results of a publicly available dataset and an in-house clinical dataset proved that our method outperformed existing semi-supervised methods. For the two datasets with a labeled ratio of 5%, the Dice similarity coefficient scores were 84.94% ± 4.31 and 81.24% ± 2.4, the HD95 values were 4.63 mm ± 2.13 and 3.94 mm ± 2.72, and the Jaccard similarity coefficient scores were 74.00% ± 6.20 and 68.49% ± 3.39, respectively. Significance. The proposed model effectively addresses the challenges of limited data samples and high costs associated with manual annotation in the medical field, leading to enhanced segmentation accuracy.

Clinical Implementation and Evaluation of Auto-Segmentation Tools for Multi-Site Contouring in Radiotherapy

Background and purposeTools for auto-segmentation in radiotherapy are widely available, but guidelines for clinical implementation are missing. The goal was to develop a workflow for performance evaluation of three commercial auto-segmentation tools to select one candidate for clinical implementation. Materials and MethodsOne hundred patients with six treatment sites (brain, head-and-neck, thorax, abdomen, and pelvis) were included. Three sets of AI-based contours for organs-at-risk (OAR) generated by three software tools and manually drawn expert contours were blindly rated for contouring accuracy. The dice similarity coefficient (DSC), the Hausdorff distance, and a dose/volume evaluation based on the recalculation of the original treatment plan were assessed. Statistically significant differences were tested using the Kruskal-Wallis test and the post-hoc Dunn Test with Bonferroni correction. ResultsThe mean DSC scores compared to expert contours for all OARs combined were 0.80 ± 0.10, 0.75 ± 0.10, and 0.74 ± 0.11 for the three software tools. Physicians' rating identified equivalent or superior performance of some AI-based contours in head (eye, lens, optic nerve, brain, chiasm), thorax (e.g., heart and lungs), and pelvis and abdomen (e.g., kidney, femoral head) compared to manual contours. For some OARs, the AI models provided results requiring only minor corrections. Bowel-bag and stomach were not fit for direct use. During the interdisciplinary discussion, the physicians' rating was considered the most relevant. ConclusionA comprehensive method for evaluation and clinical implementation of commercially available auto-segmentation software was developed. The in-depth analysis yielded clear instructions for clinical use within the radiotherapy department.