Dice Index Research Articles

Markerless tracking of tumor and tissues: A motion model approach.

Respiratory motion management is essential in order to achieve high-precision radiotherapy. Markerless motion tracking of tumor can provide a non-invasive way to manage respiratory motion, thereby enhancing treatment accuracy. However, the low contrast in real-time x-ray images for image guidance limits the application of markerless tracking. We present a novel approach based on a motion model to perform markerless tracking of tumor and surrounding tissues even when they have low contrast in real-time x-ray images. A proof-of-concept validation of the method has been performed using digital and physical phantoms at breathing conditions that are significantly different than the planning stage. A motion model is first constructed by performing principal component analysis (PCA) on the planning 4DCT. During treatment, the motion of a surrogate is tracked and used as the input of the motion model, which generates a 3D real-time volume estimation. Such 3D estimation is then projected to 2D to create digitally reconstructed radiographs (DRRs). The relationships between the real-time DRRs, reference DRRs, and reference x-ray images are first established to simulate 2D real-time images from the real-time volume. The registration between the simulated 2D real-time images and real-time x-ray images corrects the initial motion model estimation to ensure the estimated volume matches the real-time condition. In digital phantom, the Dice index of pancreas was improved from 0.74 to 0.78 after correction using real-time DRRs in fully inhaled phase. Validation on lung and pancreas is performed in physical phantom with two motion traces. The surrogate-tumor relationships were intentionally altered to generate large target localization errors due to the differences in body condition between treatment planning stage and during treatment. The real-time correction for the estimated 3D real-time volume was performed using a pair of 2D x-ray images. For the deep breathing motion trace, the tumor localization mean absolute error (MAE) throughout the tracking decreases from around 3mm to less than 1mm after correction. For the shallow breathing motion trace with a 1.7mm baseline shift, the tumor localization MAE throughout the tracking decreases from around 1.5mm to less than 1mm after correction. The method combines the detailed structural information from planning 4DCT and real-time information from real-time x-ray images through a motion model. The matching between the real-time model estimation and 2D real-time images is performed in the same modality so that it can be applied to regions with low contrast in the images. The real-time images successfully corrected the initial motion model estimations in our proof-of-concept validation. This suggests the potential to perform markerless tracking in low-contrast region using a motion model.

Investigations of Acute Ischemic Stroke Detection Techniques from CT Images Using Machine Learning Approaches

Stroke is a kind of cerebrovascular accident (CVA) that primarily affects individuals over the age of 50, while it can afflict people of all ages. Stroke patients suffer from a chronic condition that leaves them physically unable till their death. To identify the onset of this illness, numerous studies have been carried out. Common signs of a stroke include stiffness, a shift in posture, and tremors in the arms and legs. Functional and emotional mediation are linked to a significant area of the brain. The symptoms of many brain disorders are similar in the early stages because of disruptions to dopamine and other regulating mechanisms. This is a crucial step in how stroke lesions develop. The CAD system is the most efficient method for identifying a stroke. By speeding up the analysis of the required aberration, the CAD system improves the process of illness diagnosis. The goal of the proposed study is to create a compact system that can lessen processing errors, false positive rates, and complexity—three major problems with the current system. The proposed study uses a machine learning model to interpret and classify a CT brain image that shows an Acute Ischaemic Stroke (AIS) lesion. To distinguish between normal and AIS stroke lesions, the pre-processed image is segmented and classed using a Random Forest classifier. Performance criteria like Peak signal ratio, average gradient, accuracy, specificity, sensitivity, and dice index are used to evaluate the experimental output of the proposed work. The suggested model has a straightforward processing structure and the highest efficiency.

NIMG-06. AUTOMATED VOLUMETRY OF MENINGIOMAS IN CONTRAST-ENHANCED T1-WEIGHTED MRI USING DEEP LEARNING

Abstract BACKGROUND With the widespread use of MRI equipment and brain docks, there are more opportunities to perform follow-up observation for meningiomas. In recent years, attempts have been made to predict prognosis by using tumor volume as a highly objective imaging biomarker. On the other hand, manual lesion extraction by a reader with anatomical knowledge is indispensable for volume measurement, which requires a certain amount of labor and time. This is a major barrier to the generation of evidence and clinical application in this field. OBJECTIVE To establish the clinical significance of tumor volume as an imaging biomarker in meningiomas, and to develop a simple and highly reproducible application for automated volume measurement using artificial intelligence, and to investigate its accuracy and potential for clinical application. METHODS 178 patients with meningiomas who underwent MRI at Osaka University Hospital were included in the study. Contrast-enhanced T1-weighted images were used for manual lesion extraction by two neurosurgeons. These annotated image data were randomly divided into training (90%) and testing (10%). The deeplabV3 model was used to train the artificial intelligence, and leave-one-out cross-validation was used to train and validate the automated model. The test data were used for comparison, and the agreement rate (Dice index) on the images of the automatically extracted regions and the error rate of the calculated tumor volume (mean square error rate: RMSPE) were used as indices of measurement performance. RESULTS The average agreement rate between the automatically extracted area and the test data was 92.3±5.1%. The error rate of tumor volume was 9.5±1.2%. CONCLUSION This application using artificial intelligence was considered to have a certain validity in the accuracy of automatic lesion extraction. This method, which enables simple tumor volume measurement, is expected to create new imaging biomarkers that will contribute to the optimization and personalization of meningioma treatment.

NIMG-81. PREDICTING TUMOR PROGRESSION WITH FIBER DENSITY-WEIGHTED WHITE MATTER PATHLENGTH MAPS IN PATIENTS WITH GLIOBLASTOMA

Abstract PURPOSE Although histopathological evidence indicates that glioma cells preferentially migrate along large white-matter bundles, standard-of-care (SOC) radiation therapy (RT) planning remains an isotropic expansion of the anatomical lesion. We hypothesize that anisotropic expansion of RT target volumes along white-matter pathways using fiber density-weighted, white-matter pathlength maps (DW-WMPLMs) derived from diffusion tensor imaging (DTI) could improve the prediction of tumor cell migration beyond surgical margins by a deep learning model compared to SOC clinical target volumes (CTVs). METHODS DTI and anatomical MRI from 118 patients newly-diagnosed with glioblastoma were retrospectively analyzed with anatomical imaging from post-surgical resection and latest progression scan before intervention. Parallel-transport tractography seeded from the pre-surgery T2-lesion was used to estimate white-matter fiber propagation. Fibers that intersected with the resection cavity were weighted by their density to control the length of expansion and create a DW-WMPLM that was used together with T2-FLAIR and T1-post-contrast images as inputs to a novel 3D-Swin-UNetR network, trained to predict the recurrence region using 4-fold cross-validation and evaluated in a holdout test-set (95/23 CV/test). New weighted-overlap, coverage, and sparing indices were introduced to evaluate coverage of the progressed lesion and healthy brain excluded. RESULTS 52% of patients had progression outside the standard 2cm-isotropic CTV margin, motivating the need for anisotropic CTVs. Our novel DW-WMPLMs had 87±20% and 83±18% overlap with contrast-enhancing and T2-lesion progression, respectively. The deep-learning model trained with DW-WMPLMs and anatomical images achieved 19% and 56% higher Dice and weighted-overlap indices compared to the isotropic 2cm-CTV(p<0.02), with less normal brain included. CONCLUSION This study demonstrates the feasibility and benefit of incorporating a novel metric of white-matter track characterization into deep learning progression prediction models of glioblastoma for RT-planning. Current work is validating findings in a prospective cohort acquired immediately prior to RT and incorporating other imaging metrics.

Machine Learning Models Based on [18F]FDG PET Radiomics for Bone Marrow Assessment in Non-Hodgkin Lymphoma

Non-Hodgkin lymphoma is a heterogeneous group of cancers that triggers bone marrow infiltration in 20–40% of cases. Bone marrow biopsy in combination with a visual assessment of [18F]FDG PET/CT images is used to assess the marrow status. Despite the potential of both techniques, they still have limitations due to the subjectivity of visual assessment. The present study aims to develop models based on bone marrow uptake in [18F]FDG PET/CT images at the time of diagnosis to differentiate bone marrow status. For this purpose, a model trained for skeleton segmentation and based on the U-Net architecture is retrained for bone marrow segmentation from CT images. The mask obtained from this segmentation together with the [18F]FDG PET image is used to extract radiomics features with which 11 machine learning models for marrow status differentiation are trained. The segmentation model yields very satisfactory results with Jaccard and Dice index values of 0.933 and 0.964, respectively. As for the classification models, a maximum F1_score_weighted and F1_score_macro of 0.962 and 0.747, respectively, are achieved. This highlights the potential of these features for bone marrow assessment, laying the foundation for a new clinical decision support system.

Breast tumor segmentation in ultrasound using distance-adapted fuzzy connectedness, convolutional neural network, and active contour

This study addresses computer-aided breast cancer diagnosis through a hybrid framework for breast tumor segmentation in ultrasound images. The core of the three-stage method is based on the autoencoder convolutional neural network. In the first stage, we prepare a hybrid pseudo-color image through multiple instances of fuzzy connectedness analysis with a novel distance-adapted fuzzy affinity. We produce different weight combinations to determine connectivity maps driven by particular image specifics. After the hybrid image is processed by the deep network, we adjust the segmentation outcome with the Chan-Vese active contour model. We find the idea of incorporating fuzzy connectedness into the input data preparation for deep-learning image analysis our main contribution to the study. The method is trained and validated using a combined dataset of 993 breast ultrasound images from three public collections frequently used in recent studies on breast tumor segmentation. The experiments address essential settings and hyperparameters of the method, e.g., the network architecture, input image size, and active contour setup. The tumor segmentation reaches a median Dice index of 0.86 (mean at 0.79) over the combined database. We refer our results to the most recent state-of-the-art from 2022–2023 using the same datasets, finding our model comparable in segmentation performance.

Super-resolution reconstruction of time-resolved four-dimensional computed tomography (TR-4DCT) with multiple breathing cycles based on TR-4DMRI.

Respiratory motion irregularities in lung cancer patients are common and can be severe during multi-fractional (∼20 mins/fraction) radiotherapy. However, the current clinical standard of motion management is to use a single-breath respiratory-correlated four-dimension computed tomography (RC-4DCT or 4DCT) to estimate tumor motion to delineate the internal tumor volume (ITV), covering the trajectory of tumor motion, as a treatment target. To develop a novel multi-breath time-resolved (TR) 4DCT using the super-resolution reconstruction framework with TR 4D magnetic resonance imaging (TR-4DMRI) as guidance for patient-specific breathing irregularity assessment, overcoming the shortcomings of RC-4DCT, including binning artifacts and single-breath limitations. Six lung cancer patients participated in the IRB-approved protocol study to receive multiple T1w MRI scans, besides an RC-4DCT scan on the simulation day, including 80 low-resolution (lowR: 5×5×5mm3) free-breathing (FB) 3D cine MRFB images in 40s (2Hz) and a high-resolution (highR: 2×2×2mm3) 3D breath-hold (BH) MRBH image for each patient. A CT (1×1×3mm3) image was selected from 10-bin RC-4DCT with minimal binning artifacts and a close diaphragm match (<1cm) to the MRBH image. A mutual-information-based Freeform deformable image registration (DIR) was used to register the CT and MRBH via the opposite directions (namely F1: and F2: ) to establish CT-MR voxel correspondences. An intensity-based enhanced Demons DIR was then applied for , in which the original MRBH was used in D1: , while the deformed MRBH was used in D2: . The deformation vector fields (DVFs) obtained from each DIR were composed to apply to the deformed CT (D1) and original CT (D2) to reconstruct TR-4DCT images. A digital 4D-XCAT phantom at the end of inhalation (EOI) and end of exhalation (EOE) with 2.5cm diaphragmatic motion and three spherical targets (ϕ=2, 3, 4cm) were first tested to reconstruct TR-4DCT. For each of the six patients, TR-4DCT images at the EOI, middle (MID), and EOE were reconstructed with both D1 and D2 approaches. TR-4DCT image quality was evaluated with mean distance-to-agreement (MDA) at the diaphragm compared with MRFB, tumor volume ratio (TVR) referenced to MRBH, and tumor shape difference (DICE index) compared with the selected input CT. Additionally, differences in the tumor center of mass (|∆COMD1-D2|), together with TVR and DICE comparison, was assessed in the D1 and D2 reconstructed TR-4DCT images. In the phantom, TR-4DCT quality is assessed by MDA=2.0±0.8mm at the diaphragm, TVR=0.8±0.0 for all tumors, and DICE=0.83±0.01, 0.85±0.02, 0.88±0.01 for ϕ=2, 3, 4cm tumors, respectively. In six patients, the MDA in diaphragm match is -1.6±3.1mm (D1) and 1.0±3.9mm (D2) between the reconstructed TR-4DCT and lowR MRFB among 18 images (3 phases/patient). The tumor similarity is TVR=1.2±0.2 and DICE=0.70±0.07 for D1 and TVR=1.4±0.3 (D2) and DICE=0.73±0.07 for D2. The tumor position difference is |∆COMD1-D2|=1.2±0.8mm between D1 and D2 reconstructions. The feasibility of super-resolution reconstruction of multi-breathing-cycle TR-4DCT is demonstrated and image quality at the diaphragm and tumor is assessed in both the 4D-XCAT phantom and six lung cancer patients. The similarity of D1 and D2 reconstruction suggests consistent and reliable DIR results. Clinically, TR-4DCT has the potential for breathing irregularity assessment and dosimetry evaluation in radiotherapy.

Reinforcement learning-based method for type B aortic dissection localization

In the segmentation of aortic dissection, there are issues such as low contrast between the aortic dissection and surrounding organs and vessels, significant differences in dissection morphology, and high background noise. To address these issues, this paper proposed a reinforcement learning-based method for type B aortic dissection localization. With the assistance of a two-stage segmentation model, the deep reinforcement learning was utilized to perform the first-stage aortic dissection localization task, ensuring the integrity of the localization target. In the second stage, the coarse segmentation results from the first stage were used as input to obtain refined segmentation results. To improve the recall rate of the first-stage segmentation results and include the segmentation target more completely in the localization results, this paper designed a reinforcement learning reward function based on the direction of recall changes. Additionally, the localization window was separated from the field of view window to reduce the occurrence of segmentation target loss. Unet, TransUnet, SwinUnet, and MT-Unet were selected as benchmark segmentation models. Through experiments, it was verified that the majority of the metrics in the two-stage segmentation process of this paper performed better than the benchmark results. Specifically, the Dice index improved by 1.34%, 0.89%, 27.66%, and 7.37% for each respective model. In conclusion, by incorporating the type B aortic dissection localization method proposed in this paper into the segmentation process, the overall segmentation accuracy is improved compared to the benchmark models. The improvement is particularly significant for models with poorer segmentation performance.

Deep learning-based segmentation of abdominal aortic aneurysms and intraluminal thrombus in 3D ultrasound images.

Ultrasound (US)-based patient-specific rupture risk analysis of abdominal aortic aneurysms (AAAs) has shown promising results. Input for these models is the patient-specific geometry of the AAA. However, segmentation of the intraluminal thrombus (ILT) remains challenging in US images due to the low ILT-blood contrast. This study aims to improve AAA and ILT segmentation in time-resolved three-dimensional (3D + t) US images using a deep learning approach. In this study a "no new net" (nnU-Net) model was trained on 3D + t US data using either US-based or (co-registered) computed tomography (CT)-based annotations. The optimal training strategy for this low-contrast data was determined for a limited dataset. The merit of augmentation was investigated, as well as the inclusion of low-contrast areas. Segmentation results were validated with CT-based geometries as the ground truth. The model trained on CT-based masks showed the best performance in terms of DICE index, Hausdorff distance, and diameter differences, covering a larger part of the AAA. With a higher accuracy and less manual input the model outperforms conventional methods, with a mean Hausdorff distance of 4.4mm for the vessel and 7.8mm for the lumen. However, visibility of the lumen-ILT interface remains the limiting factor, necessitating improvements in image acquisition to ensure broader patient inclusion and enable rupture risk assessment of AAAs in the future.

Prospective Longitudinal Analysis of Physiologic MRI-based Tumor Habitat Predicts Short-term Patient Outcomes in IDH-wildtype Glioblastoma.

This study validates MRI-based tumor habitats in predicting time-to-progression (TTP), overall survival (OS), and progression site in isocitrate dehydrogenase (IDH)-wildtype glioblastoma patients. Seventy-nine patients were prospectively enrolled between January 2020 and June 2022. MRI, including diffusion-weighted and dynamic susceptibility contrast imaging, were obtained immediately post-operation and at three serial timepoints. Voxels from cerebral blood volume and apparent diffusion coefficient maps were grouped into three habitats (hypervascular cellular, hypovascular cellular, and nonviable tissue) using k-means clustering. Pre-defined cutoffs for increases in hypervascular and hypovascular cellular habitat were applied to calculate the habitat risk score. Associations between spatiotemporal habitats, habitat risk score, TTP, and OS were investigated using Cox proportional hazards modeling. Habitat risk score was compared to tumor volume using time-dependent receiver operating characteristics analysis. Progression sites were matched with spatial habitats. Increases in hypervascular and hypovascular cellular habitats and habitat risk scores were associated with shorter TTP and OS (all p < .05). Hypovascular cellular habitat and habitat risk scores 1 and 2 independently predicted TTP (hazard ratio [HR], 4.14; p=.03, HR, 4.51; p=.001 and HR, 10.02; p<.001, respectively). Hypovascular cellular habitat and habitat risk score 2 independently predicted OS (HR, 4.01, p=.003; and HR, 3.27, p<.001, respectively). Habitat risk score outperformed tumor volume in predicting TTP (12-month AUC, 0.762 vs. 0.646, p=.048). Hypovascular cellular habitat predicted progression sites (mean Dice index: 0.31). Multiparametric physiologic MRI-based spatiotemporal tumor habitats and habitat risk scores are useful biomarkers for early tumor progression and outcomes in IDH-wildtype glioblastoma patients.

Colorectal cancer histopathology image analysis: A comparative study of prognostic values of automatically extracted morphometric nuclear features in multispectral and red-blue-green imagery.

Multispectral imaging (MSI) has been utilized to predict the prognosis of colorectal cancer (CRC) patients, however, our understanding of the prognostic value of nuclear morphological parameters of bright-field MSI in CRC is still limited. This study was designed to compare the efficiency of MSI and standard red-green-blue (RGB) images in predicting the prognosis of CRC. We compared the efficiency of MS and conventional RGB images on the quantitative assessment of hematoxylin-eosin (HE) stained histopathology images. A pipeline was developed using a pixel-wise support vector machine (SVM) classifier for gland-stroma segmentation, and a marker-controlled watershed algorithm was used for nuclei segmentation. The correlation between extracted morphological parameters and the five-year disease-free survival (5-DFS) was analyzed. Forty-seven nuclear morphological parameters were extracted in total. Based on Kaplan-Meier analysis, eight features derived from MS images and seven featured derived from RGB images were significantly associated with 5-DFS, respectively. Compared with RGB images, MSI showed higher accuracy, precision, and Dice index in nuclei segmentation. Multivariate analysis indicated that both integrated parameters 1 (factors negatively correlated with CRC prognosis including nuclear number, circularity, eccentricity, major axis length) and 2 (factors positively correlated with CRC prognosis including nuclear average area, area perimeter, total area/total perimeter ratio, average area/perimeter ratio) in MS images were independent prognostic factors of 5-DFS, in contrast with only integrated parameter 1 (P<0.001) in RGB images. More importantly, the quantification of HE-stained MS images displayed higher accuracy in predicting 5-DFS compared with RGB images (76.9% vs 70.9%). Quantitative evaluation of HE-stained MS images could yield more information and better predictive performance for CRC prognosis than conventional RGB images, thereby contributing to precision oncology.

The Development of a Yolov8-Based Model for the Measurement of Critical Shoulder Angle (CSA), Lateral Acromion Angle (LAA), and Acromion Index (AI) from Shoulder X-ray Images.

Background: The accurate and effective evaluation of parameters such as critical shoulder angle, lateral acromion angle, and acromion index from shoulder X-ray images is crucial for identifying pathological changes and assessing disease risk in the shoulder joint. Methods: In this study, a YOLOv8-based model was developed to automatically measure these three parameters together, contributing to the existing literature. Initially, YOLOv8 was used to segment the acromion, glenoid, and humerus regions, after which the CSA, LAA angles, and AI between these regions were calculated. The MURA dataset was employed in this study. Results: Segmentation performance was evaluated with the Dice and Jaccard similarity indices, both exceeding 0.9. Statistical analyses of the measurement performance, including Pearson correlation coefficient, RMSE, and ICC values demonstrated that the proposed model exhibits high consistency and similarity with manual measurements. Conclusions: The results indicate that automatic measurement methods align with manual measurements with high accuracy and offer an effective alternative for clinical applications. This study provides valuable insights for the early diagnosis and management of shoulder diseases and makes a significant contribution to existing measurement methods.

Detection and Segmentation of Mouth Region in Stereo Stream Using YOLOv6 and DeepLab v3+ Models for Computer-Aided Speech Diagnosis in Children

This paper describes a multistage framework for face image analysis in computer-aided speech diagnosis and therapy. Multimodal data processing frameworks have become a significant factor in supporting speech disorders’ treatment. Synchronous and asynchronous remote speech therapy approaches can use audio and video analysis of articulation to deliver robust indicators of disordered speech. Accurate segmentation of articulators in video frames is a vital step in this agenda. We use a dedicated data acquisition system to capture the stereovision stream during speech therapy examination in children. Our goal is to detect and accurately segment four objects in the mouth area (lips, teeth, tongue, and whole mouth) during relaxed speech and speech therapy exercises. Our database contains 17,913 frames from 76 preschool children. We apply a sequence of procedures employing artificial intelligence. For detection, we train the YOLOv6 (you only look once) model to catch each of the three objects under consideration. Then, we prepare the DeepLab v3+ segmentation model in a semi-supervised training mode. As preparation of reliable expert annotations is exhausting in video labeling, we first train the network using weak labels produced by initial segmentation based on the distance-regularized level set evolution over fuzzified images. Next, we fine-tune the model using a portion of manual ground-truth delineations. Each stage is thoroughly assessed using the independent test subset. The lips are detected almost perfectly (average precision and F1 score of 0.999), whereas the segmentation Dice index exceeds 0.83 in each articulator, with a top result of 0.95 in the whole mouth.

Segmentation and Estimation of Fetal Biometric Parameters using an Attention Gate Double U-Net with Guided Decoder Architecture

The fetus's health is evaluated with the biometric parameters obtained from the low-resolution ultrasound images. The accuracy of biometric parameters in existing protocols typically depends on conventional image processing approaches and hence, is prone to error. This study introduces the Attention Gate Double U-Net with Guided Decoder (ADU-GD) model specifically crafted for fetal biometric parameter prediction. The attention network and guided decoder are specifically designed to dynamically merge local features with their global dependencies, enhancing the precision of parameter estimation. The ADU-GD displays superior performance with Mean Absolute Error of 0.99 mm and segmentation accuracy of 99.1 % when benchmarked against the well-established models. The proposed model consistently achieved a high Dice index score of about 99.1 ± 0.8, with a minimal Hausdorff distance of about 1.01 ± 1.07 and a low Average Symmetric Surface Distance of about 0.25 ± 0.21, demonstrating the model's excellence. In a comprehensive evaluation, ADU-GD emerged as a frontrunner, outperforming existing deep-learning models such as Double U-Net, DeepLabv3, FCN-32s, PSPNet, SegNet, Trans U-Net, Swin U-Net, Mask-R2CNN, and RDHCformer models in terms of Mean Absolute Error for crucial fetal dimensions, including Head Circumference, Abdomen Circumference, Femur Length, and BiParietal Diameter. It achieved superior accuracy with MAE values of 2.2 mm, 2.6 mm, 0.6 mm, and 1.2 mm, respectively.

Accuracy of thoracic nerves recognition for surgical support system using artificial intelligence

We developed a surgical support system that visualises important microanatomies using artificial intelligence (AI). This study evaluated its accuracy in recognising the thoracic nerves during lung cancer surgery. Recognition models were created with deep learning using images precisely annotated for nerves. Computational evaluation was performed using the Dice index and the Jaccard index. Four general thoracic surgeons evaluated the accuracy of nerve recognition. Further, the differences in time lag, image quality and smoothness of movement between the AI system and surgical monitor were assessed. Ratings were made using a five-point scale. The computational evaluation was relatively favourable, with a Dice index of 0.56 and a Jaccard index of 0.39. The AI system was used for 10 thoracoscopic surgeries for lung cancer. The accuracy of thoracic nerve recognition was satisfactory, with a recall score of 4.5 ± 0.4 and a precision score of 4.0 ± 0.9. Though smoothness of motion (3.2 ± 0.4) differed slightly, nearly no difference in time lag (4.9 ± 0.3) and image quality (4.6 ± 0.5) between the AI system and the surgical monitor were observed. In conclusion, the AI surgical support system has a satisfactory accuracy in recognising the thoracic nerves.

Segmentation of stroke lesions using transformers-augmented MRI analysis.

Accurate segmentation of chronic stroke lesions from mono-spectral magnetic resonance imaging scans (e.g., T1-weighted images) is a difficult task due to the arbitrary shape, complex texture, variable size and intensities, and varied locations of the lesions. Due to this inherent spatial heterogeneity, existing machine learning methods have shown moderate performance for chronic lesion delineation. In this study, we introduced: (1) a method that integrates transformers' deformable feature attention mechanism with convolutional deep learning architecture to improve the accuracy and generalizability of stroke lesion segmentation, and (2) an ecological data augmentation technique based on inserting real lesions into intact brain regions. Our combination of these two approaches resulted in a significant increase in segmentation performance, with a Dice index of 0.82 (±0.39), outperforming the existing methods trained and tested on the same Anatomical Tracings of Lesions After Stroke (ATLAS) 2022 dataset. Our method performed relatively well even for cases with small stroke lesions. We validated the robustness of our method through an ablation study and by testing it on new unseen brain scans from the Ischemic Stroke Lesion Segmentation (ISLES) 2015 dataset. Overall, our proposed approach of transformers with ecological data augmentation offers a robust way to delineate chronic stroke lesions with clinically relevant accuracy. Our method can be extended to other challenging tasks that require automated detection and segmentation of diverse brain abnormalities from clinical scans.

Continuum topological derivative - A novel application tool for segmentation of CT and MRI images

IntroductionComputed Tomography (CT) and Magnetic Resonance Imaging (MRI) are essential tools for unraveling anatomical and tissue properties, particularly in the head and brain. CT provides high-contrast images, particularly valuable in cases such as cerebral bleeds, and also aids in estimating cranial deformities and organ shape deviations. MRI, on the other hand, offers excellent imaging of cerebral artery regions, allowing analysis of various cerebral pathologies through different sequences. Beyond detecting common head and brain disorders, these modalities play a crucial role in identifying abnormalities in orbits, middle cerebral artery territories, brain ventricles, soft tissues, and bones. A unique aspect of brain MRI is its ability to produce multiplanar brain assessments. Both head/brain CT and MRI are invaluable for studying haemorrhage cases, with segmentation of affected areas providing detailed images for further analysis. This study explores the application of a novel mathematical technique, continuum topological derivative (CTD), for CT and MR image segmentation. MethodsThe initial stage of Continuum Topological Derivative (CTD) segmentation involves preprocessing CT and MR images due to their susceptibility to inherent noises, such as quantum mottle, and Gaussian and Rayleigh noises, respectively. In this study, we have implemented the CTD denoising algorithm to produce denoised CT/MR images, serving as ground truth for subsequent segmentation steps. Validation of the denoised CTD CT/MR images was conducted through minimal residual value computation across all case studies. Following this, segmentation of the region of interest was performed using the CTD technique, with comparisons made against Discrete Topological Derivatives (DTD), k-mean clustering and Adaptive Threshold methods. Evaluation of the proposed CTD algorithm's effectiveness in segmentation involved calculating performance metrics such as Jaccard and dice indices to assess spatial overlap of segmented images. ResultsThe CTD technique yields excellent segmentation results, not only for the delineated region of interest but also for volume-based cerebral blood areas and anomalies in the middle cerebral artery (MCA) and its territorial areas, which are substantiated through performance metrics and visual inspection by trained radiologist. This aids in determining the severity of stroke in affected patients. Additionally, a unique attempt is made to apply CTD to Electrical Impedance Tomography (EIT) images of the lungs for precise estimation of the breathing cycle. CTD successfully generates standardized images, demonstrating attenuation and density characteristics for cerebral cisterns, arteries, and ventricles. DiscussionThe denoised images obtained through CTD facilitate thorough analysis of both normal and pathological conditions, providing radiologists with enhanced capabilities to identify subtle details, particularly in areas such as abnormal cerebral artery territories, haemorrhage cases, cisterns, ventricles and arteries. Results clearly demonstrate that the combination of CTD denoising and segmentation outperforms the other three established methods in terms of both efficiency and accuracy in delineating diseased or affected areas, as evidenced by the various case studies conducted in this research. In summary, the proposed CTD method aims to delineate boundaries and contours of the region of interest, facilitating precise estimation of size and shape for accurate detection of the extent of diseased or affected areas.

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.

Fine grained automatic left ventricle segmentation via ROI based Tri-Convolutional neural networks.

The left ventricle segmentation (LVS) is crucial to the assessment of cardiac function. Globally, cardiovascular disease accounts for the majority of deaths, posing a significant health threat. In recent years, LVS has gained important attention due to its ability to measure vital parameters such as myocardial mass, end-diastolic volume, and ejection fraction. Medical professionals realize that manually segmenting data to evaluate these processes takes a lot of time, effort when diagnosing heart diseases. Yet, manually segmenting these images is labour-intensive and may reduce diagnostic accuracy. This paper, propose a combination of different deep neural networks for semantic segmentation of the left ventricle based on Tri-Convolutional Networks (Tri-ConvNets) to obtain highly accurate segmentation. CMRI images are initially pre-processed to remove noise artefacts and enhance image quality, then ROI-based extraction is done in three stages to accurately identify the LV. The extracted features are given as input to three different deep learning structures for segmenting the LV in an efficient way. The contour edges are processed in the standard ConvNet, the contour points are processed using Fully ConvNet and finally the noise free images are converted into patches to perform pixel-wise operations in ConvNets. The proposed Tri-ConvNets model achieves the Jaccard indices of 0.9491 ± 0.0188 for the sunny brook dataset and 0.9497 ± 0.0237 for the York dataset, and the dice index of 0.9419 ± 0.0178 for the ACDC dataset and 0.9414 ± 0.0247 for LVSC dataset respectively. The experimental results also reveal that the proposed Tri-ConvNets model is faster and requires minimal resources compared to state-of-the-art models.

Automation of Wilms’ tumor segmentation by artificial intelligence

Background3D reconstruction of Wilms’ tumor provides several advantages but are not systematically performed because manual segmentation is extremely time-consuming. The objective of our study was to develop an artificial intelligence tool to automate the segmentation of tumors and kidneys in children.MethodsA manual segmentation was carried out by two experts on 14 CT scans. Then, the segmentation of Wilms’ tumor and neoplastic kidney was automatically performed using the CNN U-Net and the same CNN U-Net trained according to the OV2ASSION method. The time saving for the expert was estimated depending on the number of sections automatically segmented.ResultsWhen segmentations were performed manually by two experts, the inter-individual variability resulted in a Dice index of 0.95 for tumor and 0.87 for kidney. Fully automatic segmentation with the CNN U-Net yielded a poor Dice index of 0.69 for Wilms’ tumor and 0.27 for kidney. With the OV2ASSION method, the Dice index varied depending on the number of manually segmented sections. For the segmentation of the Wilms’ tumor and neoplastic kidney, it varied respectively from 0.97 to 0.94 for a gap of 1 (2 out of 3 sections performed manually) to 0.94 and 0.86 for a gap of 10 (1 section out of 6 performed manually).ConclusionFully automated segmentation remains a challenge in the field of medical image processing. Although it is possible to use already developed neural networks, such as U-Net, we found that the results obtained were not satisfactory for segmentation of neoplastic kidneys or Wilms’ tumors in children. We developed an innovative CNN U-Net training method that makes it possible to segment the kidney and its tumor with the same precision as an expert while reducing their intervention time by 80%.