Dice Similarity Coefficient Research Articles

Advancing precision in medical image segmentation: A performance analysis of loss functions for COVID‐19 lung infection segmentation in computed tomography images

AbstractThis study evaluates the effectiveness of three loss functions Asymmetric Unified Focal Loss (AUFL), Dice Similarity Coefficient Loss (DSCL), and Cross‐Entropy (CE) for segmenting COVID‐19 lung infections in computed tomography images. Detailed analyses using the intersection over union metric assessed each function's accuracy. AUFL achieved an average Dice Similarity Coefficient (DSC) of 85.18% ± 8.86%, outperforming DSCL 85.18% ± 8.86%, which had the same average DSC but less precise segmentation, and CE, which had an average DSC of 78.31% ± 11.93%. Segmentations using AUFL demonstrated more defined contours and better alignment with actual anatomical structures than those obtained with DSCL and CE. Observations revealed that AUFL‐generated segmentations had more precise boundaries and were more consistent with the expected anatomical regions of lung infections. This study is the first to quantitatively and qualitatively compare the effectiveness of AUFL, DSCL, and CE in segmenting COVID‐19 lung infections, providing concrete evidence of AUFL's superiority in segmentation performance and reliability for clinical applications. The findings underscore the importance of selecting appropriate loss functions to enhance segmentation in medical imaging, highlighting their crucial role in improving image‐based diagnostics and treatment. The study emphasizes the need for ongoing research to optimize these segmentation techniques further.

Automatic Segmentation of Ultrasound-Guided Quadratus Lumborum Blocks Based on Artificial Intelligence.

Ultrasound-guided quadratus lumborum block (QLB) technology has become a widely used perioperative analgesia method during abdominal and pelvic surgeries. Due to the anatomical complexity and individual variability of the quadratus lumborum muscle (QLM) on ultrasound images, nerve blocks heavily rely on anesthesiologist experience. Therefore, using artificial intelligence (AI) to identify different tissue regions in ultrasound images is crucial. In our study, we retrospectively collected 112 patients (3162 images) and developed a deep learning model named Q-VUM, which is a U-shaped network based on the Visual Geometry Group 16 (VGG16) network. Q-VUM precisely segments various tissues, including the QLM, the external oblique muscle, the internal oblique muscle, the transversus abdominis muscle (collectively referred to as the EIT), and the bones. Furthermore, we evaluated Q-VUM. Our model demonstrated robust performance, achieving mean intersection over union (mIoU), mean pixel accuracy, dice coefficient, and accuracy values of 0.734, 0.829, 0.841, and 0.944, respectively. The IoU, recall, precision, and dice coefficient achieved for the QLM were 0.711, 0.813, 0.850, and 0.831, respectively. Additionally, the Q-VUM predictions showed that 85% of the pixels in the blocked area fell within the actual blocked area. Finally, our model exhibited stronger segmentation performance than did the common deep learning segmentation networks (0.734 vs. 0.720 and 0.720, respectively). In summary, we proposed a model named Q-VUM that can accurately identify the anatomical structure of the quadratus lumborum in real time. This model aids anesthesiologists in precisely locating the nerve block site, thereby reducing potential complications and enhancing the effectiveness of nerve block procedures.

Artificial intelligence based real time colorectal cancer screening study: Polyp segmentation and classification using multi-house database

PurposeThis work explores the potential of artificial intelligence (AI) for real-time colorectal cancer (CRC) screening. We utilized a multi-house dataset to achieve a three-step approach (i) automated polyp segmentation, (ii) analysis of machine learning (ML) for classification and (iii) to compare deep learning (DL) exploration for automated classification. MethodsTo achieve our objectives, we leveraged the CRPU-Net architecture for automated polyp segmentation within the multi-house colonoscopy images. Subsequently, a ML framework was employed for binary classification of the segmented regions. Feature extraction was performed using SIFT and ORB descriptors, followed by the feature fusion and dimensionality reduction through Principal Component Analysis (PCA) to create an optimal feature representation for classification. We further explored the potential of DL by investigating the use of SOTA and vision transformer (ViT) model. Additionally, we considered the development of a hybrid model (CRP-ViT) by potentially combining the ViT model with DL architectures. ResultsCRPU-Net achieved outstanding segmentation performance of 96.56% accuracy, 95.40% IoU, and 97.39% Dice coefficient. Additionally, the CRP-ViT model demonstrated promising results in binary classification of 96.59% accuracy, 96.38% sensitivity, 96.80% specificity, 96.82% precision, and 96.36% NPV. ConclusionThis study implemented CRPU-Net and CRP-ViT, AI systems designed for real-time polyp segmentation, and potential precancerous polyp classification. This comprehensive approach offers the potential to expedite colonoscopies, minimize unnecessary biopsies, improve patient care, and optimize healthcare resource allocation.

Cross-Modality Image Translation From Brain 18 F-FDG PET/CT Images to Fluid-Attenuated Inversion Recovery Images Using the CypixGAN Framework.

PET/CT and MRI can accurately diagnose dementia but are expensive and inconvenient for patients. Therefore, we aimed to generate synthetic fluid-attenuated inversion recovery (FLAIR) images from 18 F-FDG PET and CT images of the human brain using a generative adversarial network (GAN)-based deep learning framework called the CypixGAN, which combined the CycleGAN framework with the L1 loss function of the pix2pix. Data from 143 patients who underwent PET/CT and MRI were used for training (n = 79), validation (n = 20), and testing (n = 44) the deep learning frameworks. Synthetic FLAIR images were generated using the pix2pix, CycleGAN, and CypixGAN, and white matter hyperintensities (WMHs) were then segmented. The performance of CypixGAN was compared with that of the other frameworks. The CypixGAN outperformed the pix2pix and CycleGAN in generating synthetic FLAIR images with superior visual quality. Peak signal-to-noise ratio and structural similarity index (mean ± standard deviation) estimated using the CypixGAN (20.23 ± 1.31 and 0.80 ± 0.02, respectively) were significantly higher than those estimated using the pix2pix (19.35 ± 1.43 and 0.79 ± 0.02, respectively) and CycleGAN (18.74 ± 1.49 and 0.78 ± 0.02, respectively) ( P < 0.001). WMHs in synthetic FLAIR images generated using the CypixGAN closely resembled those in ground-truth images, as indicated by the low absolute percentage volume differences and high dice similarity coefficients. The CypixGAN generated high-quality FLAIR images owing to the preservation of spatial information despite using unpaired images. This framework may help improve diagnostic performance and cost-effectiveness of PET/CT when MRI scan is unavailable.

A clustering tool for generating biological geometries for computational modeling in radiobiology.

To develop a computational tool that converts biological images into geometries compatible with computational software dedicated to the Monte Carlo simulation of radiation transport (TOPAS), and subsequent biological tissue responses (CompuCell3D). The depiction of individual biological entities from segmentation images is essential in computational radiobiological modeling for two reasons: image pixels or voxels representing a biological structure, like a cell, should behave as a single entity when simulating biological processes, and the action of radiation in tissues is described by the association of biological endpoints to physical quantities, as radiation dose, scored the entire group of voxels assembling a cell.&#xD;Approach: The tool is capable of cropping and resizing the images and performing clustering of image voxels to create independent entities (clusters) by assigning a unique identifier to these voxels conforming to the same cluster. The clustering algorithm is based on the adjacency of voxels with image values above an intensity threshold to others already assigned to a cluster. The performance of the tool to generate geometries that reproduced original images was evaluated by the Dice Similarity Coefficient (DSC), and by the number of individual entities in both geometries. A set of tests consisting of segmentation images of cultured neuroblastoma cells, two cell nucleus populations, and the vasculature of a mouse brain were used.&#xD;Main results: The DSC was 1.0 in all images, indicating that original and generated geometries were identical, and the number of individual entities in both geometries agreed, proving the ability of the tool to cluster voxels effectively following user-defined specifications. The potential of this tool in computational radiobiological modeling, was shown by evaluating the spatial distribution of DNA Double-Strand-Breaks after microbeam irradiation in a segmentation image of a cell culture.&#xD;Significance: This tool enables the use of realistic biological geometries in computational radiobiological studies.

OAR-UNet: Enhancing Long-Distance Dependencies for Head and Neck OAR Segmentation

Accurate segmentation of organs at risk (OARs) is a crucial step in the precise planning of radiotherapy for head and neck tumors. However, manual segmentation methods using CT images, which are still predominantly applied in clinical settings, are inefficient and expensive. Additionally, existing segmentation methods struggle with small organs and have difficulty managing the complex interdependencies between organs. To address these issues, this study proposed an OAR-UNet segmentation method based on a U-shaped architecture with two key designs. To tackle the challenge of segmenting small organs, a Local Feature Perception Module (LFPM) is developed to enhance the sensitivity of the method to subtle structures. Furthermore, a Cross-shaped Transformer Block (CSTB) with a cross-shaped attention mechanism is introduced to improve the ability of the model to capture and process long-distance dependency information. To accelerate the convergence of the Transformer, we designed a Local Encoding Module (LEM) based on depthwise separable convolutions. In our experimental evaluation, we utilized two publicly available datasets, SegRap2023 and PDDCA, achieving Dice coefficients of 78.22% and 89.42%, respectively. These results demonstrate that our method outperforms both previous classic methods and state-of-the-art (SOTA) methods.

A Multi-Task Model for Pulmonary Nodule Segmentation and Classification.

In the computer-aided diagnosis of lung cancer, the automatic segmentation of pulmonary nodules and the classification of benign and malignant tumors are two fundamental tasks. However, deep learning models often overlook the potential benefits of task correlations in improving their respective performances, as they are typically designed for a single task only. Therefore, we propose a multi-task network (MT-Net) that integrates shared backbone architecture and a prediction distillation structure for the simultaneous segmentation and classification of pulmonary nodules. The model comprises a coarse segmentation subnetwork (Coarse Seg-net), a cooperative classification subnetwork (Class-net), and a cooperative segmentation subnetwork (Fine Seg-net). Coarse Seg-net and Fine Seg-net share identical structure, where Coarse Seg-net provides prior location information for the subsequent Fine Seg-net and Class-net, thereby boosting pulmonary nodule segmentation and classification performance. We quantitatively and qualitatively analyzed the performance of the model by using the public dataset LIDC-IDRI. Our results show that the model achieves a Dice similarity coefficient (DI) index of 83.2% for pulmonary nodule segmentation, as well as an accuracy (ACC) of 91.9% for benign and malignant pulmonary nodule classification, which is competitive with other state-of-the-art methods. The experimental results demonstrate that the performance of pulmonary nodule segmentation and classification can be improved by a unified model that leverages the potential correlation between tasks.

Optimal Training Positive Sample Size Determination for Deep Learning with a Validation on CBCT Image Caries Recognition.

Objectives: During deep learning model training, it is essential to consider the balance among the effects of sample size, actual resources, and time constraints. Single-arm objective performance criteria (OPC) was proposed to determine the optimal positive sample size for training deep learning models in caries recognition. Methods: An expected sensitivity (PT) of 0.6 and a clinically acceptable sensitivity (P0) of 0.5 were applied to the single-arm OPC calculation formula, yielding an optimal training set comprising 263 carious teeth. U-Net, YOLOv5n, and CariesDetectNet were trained and validated using clinically self-collected cone-beam computed tomography (CBCT) images that included varying quantities of carious teeth. To assess performance, an additional dataset was utilized to evaluate the accuracy of caries detection by both the models and two dental radiologists. Results: When the number of carious teeth reached approximately 250, the models reached the optimal performance levels. U-Net demonstrated superior performance, achieving accuracy, sensitivity, specificity, F1-Score, and Dice similarity coefficients of 0.9929, 0.9307, 0.9989, 0.9590, and 0.9435, respectively. The three models exhibited greater accuracy in caries recognition compared to dental radiologists. Conclusions: This study demonstrated that the positive sample size of CBCT images containing caries was predictable and could be calculated using single-arm OPC.

A hybrid PCA acceleration method for rapid real-time 2D MRI.

Objective.To develop a 2D MR acceleration method utilizing principal component analysis (PCA) in a hybrid fashion for rapid real-time applications.Approach.Retrospective testing was performed on 10 lung, 10 liver and 10 prostate 3T MRI data sets for image quality and target contourability. Sampling of k-space is performed by acquiring central (low-frequency) data in every frame while the high-frequency data is incoherently undersampled such that all of k-space is acquired in a pre-determined number of frames. Firstly, principal components (PCs) representative of intra-frame correlations between central and outer k-space data are used to estimate unsampled data in the frame of interest. Then to add further stability, PCs representative of time-domain fluctuations within a reconstruction window of the most recent frames are fit to outer k-space data (including above estimations) to obtain final estimates in the frame of interest. Accelerated reconstructions between 3x and 8x were tested for image quality and contourability along with the optimal number of PCs for fitting.Main results.It was found that at higher acceleration rates, image quality did not deteriorate significantly. Similarly, it was found that the images were of sufficient quality to contour a target using auto-contouring software at all tested acceleration rates and sites. SSIM values were found to be ⩾0.91 at all accelerations tested. Similarly dice coefficients at the different sites were found to be ⩾0.89 even at 8x accelerations which is on par with or better than intra-observer variation.Significance.This method appears to produce improved image quality and contourability compared to previous PCA methods while also allowing a greater number of PCs to be used in reconstruction. The method can be run using a simple single-channel coil and does not require significant computing power to meet real-time interventional standards (reconstruction times ∼60 ms/frame on Intel i5 CPU).

Automated abdominal organ segmentation algorithms for non-enhanced CT for volumetry and 3D radiomics analysis

Abstract Purpose To develop fully-automated abdominal organ segmentation algorithms from non-enhanced abdominal CT and low-dose chest CT and assess their feasibility for automated CT volumetry and 3D radiomics analysis of abdominal solid organs. Methods Fully-automated nnU-Net-based models were developed to segment the liver, spleen, and both kidneys in non-enhanced abdominal CT, and the liver and spleen in low-dose chest CT. 105 abdominal CTs and 60 low-dose chest CTs were used for model development, and 55 abdominal CTs and 10 low-dose chest CTs for external testing. The segmentation performance for each organ was assessed using the Dice similarity coefficients, with manual segmentation results serving as the ground truth. Agreements between ground-truth measurements and model estimates of organ volume and 3D radiomics features were assessed using the Bland–Altman analysis and intraclass correlation coefficients (ICC). Results The models accurately segmented the liver, spleen, right kidney, and left kidney in abdominal CT and the liver and spleen in low-dose chest CT, showing mean Dice similarity coefficients in the external dataset of 0.968, 0.960, 0.952, and 0.958, respectively, in abdominal CT, and 0.969 and 0.960, respectively, in low-dose chest CT. The model-estimated and ground truth volumes of these organs exhibited mean differences between − 0.7% and 2.2%, with excellent agreements. The automatically extracted mean and median Hounsfield units (ICCs, 0.970–0.999 and 0.994–0.999, respectively), uniformity (ICCs, 0.985–0.998), entropy (ICCs, 0.931–0.993), elongation (ICCs, 0.978–0.992), and flatness (ICCs, 0.973–0.997) showed excellent agreement with ground truth measurements for each organ; however, skewness (ICCs, 0.210–0.831), kurtosis (ICCs, 0.053–0.933), and sphericity (ICCs, 0.368–0.819) displayed relatively low and inconsistent agreement. Conclusion Our nnU-Net-based models accurately segmented abdominal solid organs in non-enhanced abdominal and low-dose chest CT, enabling reliable automated measurements of organ volume and specific 3D radiomics features.

Artificial intelligence-based analysis of behavior and brain images in cocaine-self-administered marmosets

BackgroundThe sophisticated behavioral and cognitive repertoires of non-human primates (NHPs) make them suitable subjects for studies involving cocaine self-administration (SA) schedules. However, ethical considerations, adherence to the 3Rs principle (replacement, reduction and refinement), and other factors make it challenging to obtain NHPs individuals for research. Consequently, there is a need for methods that can comprehensively analyze small datasets using artificial intelligence (AI). New methodsWe employed AI to identify cocaine dependence patterns from collected data. First, we collected behavioral data from cocaine SA marmosets (Callithrix jacchus) to develop a dependence prediction model. SHapley Additive exPlanations (SHAP) values were used to demonstrate the importance of various variables. Additionally, we collected positron emission tomographic (PET) images showing dopamine transporter (DAT) binding potential and developed an algorithm for PET image segmentation. ResultsThe prediction model indicated that the Random Forest (RF) algorithm performed best, with an area under the curve (AUC) of 0.92. The top five variables influencing the model were identified using SHAP values. The PET image segmentation model achieved an accuracy of 0.97, a mean squared error of 0.02, an intersection over union (IoU) of 0.845, and a Dice coefficient of 0.913. Comparison with existing methods and conclusionUtilizing data from the marmoset SA experiment, we developed an ML-based dependence prediction model and analyzed variable importance rankings using SHAP. AI-based imaging segmentation methods offer a valuable tool for evaluating DAT availability in NHPs following chronic cocaine administration.

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

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. 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. 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. 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. 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.

Intelligent diagnostic method for developmental hip dislocation

BackgroundDevelopmental dislocation of the hip joint (DDH) is a condition that severely threatens children’s healthy growth. Without timely and correct treatment, it will lead to osteoarthritis and hip dysfunction in the evolution of children.ObjectiveIt is essential to develop an intelligent model for diagnosing hip dislocation and performing accurate quantitative analysis.MethodsIn this paper, 46 cases of computed tomography (CT) images were retrospectively collected, including 19 cases of hip dislocation and 27 cases of healthy people. The experiment first uses ITK-SNAP to sketch the ilium and femoral head in the original image. Then, it uses 3D U-Net to send the label of the background, ilium, and femoral head into three channels, respectively, to realize the three-dimensional segmentation of the ilium and femoral head. Next, the extraction of the surface of the acetabulum and femoral head is performed. Subsequently, the erroneous points are eliminated, and the spherical surfaces of the acetabulum and femoral head are fitted using the least squares method. Ultimately, the spherical center distance is calculated quantitatively to predict whether the hip joint is dislocated.ResultsUnder the independent test set, the segmentation average dice coefficients of the ilium and femoral head are 89% and 93%, respectively. The spherical center distance between the acetabulum and femoral head is calculated quantitatively. If the value exceeds 10 mm, it is considered a hip dislocation. Compared with the doctor’s diagnosis, the accuracy result is 94.4%.ConclusionThis paper successfully implements a precise and automated intelligent diagnostic system for the identification of hip dislocation. Commencing with the development of a 3D segmentation algorithm for the ilium and femoral head, we further introduce a novel method that computes the spherical distance for the prediction of hip dislocation. This approach provides robust quantitative analysis, thereby facilitating more informed clinical decision-making.

Anatomical and dosimetric variations during volumetric modulated arc therapy in patients with locally advanced nasopharyngeal carcinoma after induction therapy: Implications for adaptive radiation therapy

PurposeTo investigate anatomical and dosimetric changes during volumetric modulated arc therapy (VMAT) in patients with locally advanced nasopharyngeal carcinoma (LA-NPC) after induction therapy (IT) and explore characteristics of patients with notable variations. Materials and methodsFrom July 2021 to June 2023, 60 LA-NPC patients undergoing VMAT after IT were retrospectively recruited. Adaptive computed tomography (aCT), reconstructed from weekly cone-beam computed tomography(CBCT), facilitates recontouring and planning transplantation. Volume, dice similarity coefficients, and dose to target volumes and organs at risk(OARs) on planning CT(pCT) and aCT were compared to identify changing patterns. Multivariate logistic regression was used to investigate risk factors. ResultsThe volumes of PGTVnasopharynx (PGTVp), PGTVnode (PGTVn), ipsilateral and contralateral parotid glands decreased during VMAT, with reductions of 2.25 %, 6.98 %, 20.09 % and 18.00 %, respectively, at 30 fractions from baseline (P < 0.001). After 25 fractions, D99 and D95 of PGTVn decreased by 7.94 % and 4.18 % from baseline, respectively, while the Dmean of ipsilateral and contralateral parotid glands increased by 7.80 % and 6.50 %, marking the peak rates of dosimetric variations (P < 0.001). The dosimetric fluctuations in PGTVp, the brainstem, and the spinal cord remained within acceptable limits. Furthermore, an initial BMI ≥ 23.5 kg/m2 and not-achieving objective response (OR) after IT were regarded as risk factors for a remarkable PGTVn dose reduction in the later stages of VMAT. ConclusionsReplanning for post-IT LA-NPC patients appears reasonable at 25F during VMAT. Patients with an initial BMI ≥ 23.5 kg/m2 and not-achieving OR after IT should be considered for adaptive radiation therapy to stabilize the delivered dose.

Robust unsupervised texture segmentation for motion analysis in ultrasound images.

Ultrasound imaging has emerged as a promising cost-effective and portable non-irradiant modality for the diagnosis and follow-up of diseases. Motion analysis can be performed by segmenting anatomical structures of interest before tracking them over time. However, doing so in a robust way is challenging as ultrasound images often display a low contrast and blurry boundaries. In this paper, a robust descriptor inspired from the fractal dimension is presented to locally characterize the gray-level variations of an image. This descriptor is an adaptive grid pattern whose scale locally varies as the gray-level variations of the image. Robust features are then located based on the gray-level variations, which are more likely to be consistently tracked over time despite the presence of noise. The method was validated on three datasets: segmentation of the left ventricle on simulated echocardiography (Dice coefficient, DC), accuracy of diaphragm motion tracking for healthy subjects (mean sum of distances, MSD) and for a scoliosis patient (root mean square error, RMSE). Results show that the method segments the left ventricle accurately ( ) and robustly tracks the diaphragm motion for healthy subjects ( mm) and for the scoliosis patient ( mm). This method has the potential to segment structures of interest according to their texture in an unsupervised fashion, as well as to help analyze the deformation of tissues. Possible applications are not limited to US image. The same principle could also be applied to other medical imaging modalities such as MRI or CT scans.

Accurate identification of salt domes using deep learning techniques: Transformers, generative artificial intelligence and liquid state machines

AbstractAcross various global regions abundant in oil and natural gas reserves, the presence of substantial sub‐surface salt deposits holds significant relevance. Accurate identification of salt domes becomes crucial for enterprises engaged in oil and gas exploration. Our research introduces a precise method for the automatic detection of salt domes, leveraging advanced deep learning architectures such as U‐net, transformers, artificial intelligence generative models and liquid state machines. In comparison with state‐of‐the‐art techniques, our model demonstrates superior performance, achieving a stable and validated intersection over the union metric, indicating high accuracy and robustness. Furthermore, the Dice similarity coefficient attaining underscores the model's proficiency in closely aligning with ground truth across diverse scenarios. This evaluation, conducted on 1000 seismic images, reveals that our proposed architecture is not only comparable but often surpasses existing segmentation models in effectiveness and reliability.

Nested CNN architecture for three-dimensional dose distribution prediction in tomotherapy for prostate cancer.

The hypothesis of changing network layers to increase the accuracy of dose distribution prediction, instead of expanding their dimensions, which requires complex calculations, has been considered in our study. Atotal of 137 prostate cancer patients treated with the tomotherapy technique were categorized as 80% training and validating as well as 20% testing for the nested UNet and UNet architectures. Mean absolute error (MAE) was used to measure the dosimetry indices of dose-volume histograms (DVHs), and geometry indices, including the structural similarity index measure (SSIM), dice similarity coefficient (DSC), and Jaccard similarity coefficient (JSC), were used to evaluate the isodose volume (IV) similarity prediction. To verify a statistically significant difference, the two-way statistical Wilcoxon test was used at alevel of 0.05 (p < 0.05). Use of anested UNet architecture reduced the predicted dose MAE in DVH indices. The MAE for planning target volume (PTV), bladder, rectum, and right and left femur were D98% = 1.11 ± 0.90; D98% = 2.27 ± 2.85, Dmean = 0.84 ± 0.62; D98% = 1.47 ± 12.02, Dmean = 0.77 ± 1.59; D2% = 0.65 ± 0.70, Dmean = 0.96 ± 2.82; and D2% = 1.18 ± 6.65, Dmean = 0.44 ± 1.13, respectively. Additionally, the greatest geometric similarity was observed in the mean SSIM for UNet and nested UNet (0.91 vs.0.94, respectively). The nested UNet network can be considered asuitable network due to its ability to improve the accuracy of dose distribution prediction compared to the UNet network in an acceptable time.

Structural tensor and frequency guided semi-supervised segmentation for medical images.

The method of semi-supervised semantic segmentation entails training with a limited number of labeled samples alongside many unlabeled samples, aiming to reduce dependence on pixel-level annotations. Most semi-supervised semantic segmentation methods primarily focus on sample augmentation in spatial dimensions to reduce the shortage of labeled samples. These methods tend to ignore the structural information of objects. In addition, frequency-domain information also supplies another perspective to evaluate information from images, which includes different properties compared to the spatialdomain. In this study, we attempt to answer these two questions: (1) is it helpful to provide structural information of objects in semi-supervised semantic segmentation tasks for medical images? (2) is it more effective to evaluate the segmentation performance in the frequency domain compared to the spatial domain for semi-supervised medical image segmentation? Therefore, we seek to introduce structural and frequency information to improve the performance of semi-supervised semantic segmentation for medicalimages. We present a novel structural tensor loss (STL) to guide feature learning on the spatial domain for semi-supervised semantic segmentation. Specifically, STL utilizes the structural information encoded in the tensors to enforce the consistency of objects across spatial regions, thereby promoting more robust and accurate feature extraction. Additionally, we proposed a frequency-domain alignment loss (FAL) to enable the neural networks to learn frequency-domain information across different augmented samples. It leverages the inherent patterns present in frequency-domain representations to guide the network in capturing and aligning features across diverse augmentation variations, thereby enhancing the model's robustness for the inputting variations. We conduct our experiments on three benchmark datasets, which include MRI (ACDC) for cardiac, CT (Synapse) for abdomen organs, and ultrasound image (BUSI) for breast lesion segmentation. The experimental results demonstrate that our method outperforms state-of-the-art semi-supervised approaches regarding the Dice similaritycoefficient. We find the proposed approach could improve the final performance of the semi-supervised medical image segmentation task. It will help reduce the need for medical image labels. Our code will are available at https://github.com/apple1986/STLFAL.

External validation of a deep learning model for automatic segmentation of skeletal muscle and adipose tissue on abdominal computed tomography images.

Body composition assessment using computed tomography (CT) images at the L3-level is increasingly applied in cancer research. Robust high-throughput automated segmentation is key to assess large patient cohorts and to support implementation of body composition analysis into routine clinical practice. We trained and externally validated a deep learning neural network (DLNN) to automatically segment L3-CT images. Expert-drawn segmentations of visceral and subcutaneous adipose tissue (VAT/SAT) and skeletal muscle (SM) of L3-CT-images of 3,187 patients undergoing abdominal surgery were used to train a DLNN. The external validation cohort was comprised of 2,535 patients with abdominal cancer. DLNN performance was evaluated with (geometric) Dice Similarity (DS) and Lin's Concordance Correlation Coefficient. There was a strong concordance between automatic and manual segmentations with median DS for SM, VAT, and SAT of 0.97 (interquartile range, IQR: 0.95-0.98), 0.98 (IQR: 0.95-0.98), and 0.95 (IQR: 0.92-0.97), respectively. Concordance correlations were excellent: SM 0.964 (0.959-0.968), VAT 0.998 (0.998-0.998), and SAT 0.992 (0.991-0.993). Bland-Altman metrics indicated only small and clinically insignificant systematic offsets; SM radiodensity: 0.23 hounsfield units (0.5%), SM: 1.26 cm2.m-2 (2.8%), VAT: -1.02 cm2.m-2 (1.7%), and SAT: 3.24 cm2.m-2 (4.6%). A robustly-performing and independently externally validated DLNN for automated body composition analysis was developed. CT-based body composition analysis is highly prognostic for long-term overall survival in oncology. This DLNN was succesfully trained and externally validated on several large patient cohorts and will therefore enable large scale population studies and implementation of body composition analysis into clinical practice.

Multimodal registration network with multi-scale feature-crossing.

A critical piece of information for prostate intervention and cancer treatment is provided by the complementary medical imaging modalities of ultrasound (US) and magnetic resonance imaging (MRI). Therefore, MRI-US image fusion is often required during prostate examination to provide contrast-enhanced TRUS, in which image registration is a key step in multimodal image fusion. We propose a novel multi-scale feature-crossing network for the prostate MRI-US image registration task. We designed a feature-crossing module to enhance information flow in the hidden layer by integrating intermediate features between adjacent scales. Additionally, an attention block utilizing three-dimensional convolution interacts information between channels, improving the correlation between different modal features. We used 100 cases randomly selected from The Cancer Imaging Archive (TCIA) for our experiments. A fivefold cross-validation method was applied, dividing the dataset into five subsets. Four subsets were used for training, and one for testing, repeating this process five times to ensure each subset served as the test set once. We test and evaluate our technique using fivefold cross-validation. The cross-validation trials result in a median target registration error of 2.20 mm on landmark centroids and a median Dice of 0.87 on prostate glands, both of which were better than the baseline model. In addition, the standard deviation of the dice similarity coefficient is 0.06, which suggests that the model is stable. We propose a novel multi-scale feature-crossing network for the prostate MRI-US image registration task. A random selection of 100 cases from The Cancer Imaging Archive (TCIA) was used to test and evaluate our approach using fivefold cross-validation. The experimental results showed that our method improves the registration accuracy. After registration, MRI and TURS images were more similar in structure and morphology, and the location and morphology of cancer were more accurately reflected in the images.