Ocular Structures Segmentation from Multi-sequences MRI Using 3D Unet with Fully Connected CRFs

Simple SummaryThe noninvasive differentiation of hepatocellular carcinoma (HCC) from intrahepatic cholangiocarcinoma (ICC) remains challenging. In recent years, the number of studies on the application of radiomics in liver cancer has grown dramatically. However, there have been very few studies on the differentiation of HCC from ICC based on multisequence magnetic resonance imaging (MRI) radiomics. This study aimed to investigate the efficacy of a radiomics model based on pretherapeutic fat suppression T2-weighted imaging (FS-T2WI) and dynamic-contrast-enhanced MRI (DCE-MRI) features obtained from the arterial phase (AP) and portal venous phase (PVP) for noninvasively differentiating HCC from ICC.The purpose of this study was to investigate the efficacy of magnetic resonance imaging (MRI) radiomics in differentiating hepatocellular carcinoma (HCC) from intrahepatic cholangiocarcinoma (ICC). The clinical and MRI data of 129 pathologically confirmed HCC patients and 48 ICC patients treated at the Affiliated Hospital of North Sichuan Medical College between April 2016 and December 2021 were retrospectively analyzed. The patients were randomly divided at a ratio of 7:3 into a training group of 124 patients (90 with HCC and 34 with ICC) and a validation group of 53 patients (39 with HCC and 14 with ICC). Radiomic features were extracted from axial fat suppression T2-weighted imaging (FS-T2WI) and axial arterial-phase (AP) and portal-venous-phase (PVP) dynamic-contrast-enhanced MRI (DCE-MRI) sequences, and the corresponding datasets were generated. The least absolute shrinkage and selection operator (LASSO) method was used to select the best radiomic features. Logistic regression was used to establish radiomic models for each sequence (FS-T2WI, AP and PVP models), a clinical model for optimal clinical variables (C model) and a joint radiomics model (JR model) integrating the radiomics features of all the sequences as well as a radiomics–clinical model combining optimal radiomic features and clinical risk factors (RC model). The performance of each model was evaluated using the area under the receiver operating characteristic curve (AUC). The AUCs of the FS-T2WI, AP, PVP, JR, C and RC models for distinguishing HCC from ICC were 0.693, 0.863, 0.818, 0.914, 0.936 and 0.977 in the training group and 0.690, 0.784, 0.727, 0.802, 0.860 and 0.877 in the validation group, respectively. The results of this study suggest that MRI-based radiomics may help noninvasively differentiate HCC from ICC. The model integrating the radiomics features and clinical risk factors showed a further improvement in performance.

Modified local ternary patterns technique for brain tumour segmentation and volume estimation from MRI multi-sequence scans with GPU CUDA machine

BackgroundDistant metastases is the main failure mode of nasopharyngeal carcinoma. However, early prediction of distant metastases in NPC is extremely challenging. Deep learning has made great progress in recent years. Relying on the rich data features of radiomics and the advantages of deep learning in image representation and intelligent learning, this study intends to explore and construct the metachronous single-organ metastases (MSOM) based on multimodal magnetic resonance imaging.Patients and methodsThe magnetic resonance imaging data of 186 patients with nasopharyngeal carcinoma before treatment were collected, and the gross tumor volume (GTV) and metastatic lymph nodes (GTVln) prior to treatment were defined on T1WI, T2WI, and CE-T1WI. After image normalization, the deep learning platform Python (version 3.9.12) was used in Ubuntu 20.04.1 LTS to construct automatic tumor detection and the MSOM prediction model.ResultsThere were 85 of 186 patients who had MSOM (including 32 liver metastases, 25 lung metastases, and 28 bone metastases). The median time to MSOM was 13 months after treatment (7–36 months). The patients were randomly assigned to the training set (N = 140) and validation set (N = 46). By comparison, we found that the overall performance of the automatic tumor detection model based on CE-T1WI was the best (6). The performance of automatic detection for primary tumor (GTV) and lymph node gross tumor volume (GTVln) based on the CE-T1WI model was better than that of models based on T1WI and T2WI (AP@0.5 is 59.6 and 55.6). The prediction model based on CE-T1WI for MSOM prediction achieved the best overall performance, and it obtained the largest AUC value (AUC = 0.733) in the validation set. The precision, recall, precision, and AUC of the prediction model based on CE-T1WI are 0.727, 0.533, 0.730, and 0.733 (95% CI 0.557–0.909), respectively. When clinical data were added to the deep learning prediction model, a better performance of the model could be obtained; the AUC of the integrated model based on T2WI, T1WI, and CE-T1WI were 0.719, 0.738, and 0.775, respectively. By comparing the 3-year survival of high-risk and low-risk patients based on the fusion model, we found that the 3-year DMFS of low and high MSOM risk patients were 95% and 11.4%, respectively (p < 0.001).ConclusionThe intelligent prediction model based on magnetic resonance imaging alone or combined with clinical data achieves excellent performance in automatic tumor detection and MSOM prediction for NPC patients and is worthy of clinical application.

Computer Aided Diagnosis of liver tumors from abdominal Computer Tomography (CT) images requires segmentation of tumor. Automatic segmentation of tumor from CT images is a challenging task due in part to the variation of size, shape, and position of the tumor and on the other part to the presence of other objects with the same intensity in the CT images. Therefore, it is necessary to segment the liver first so that tumor can then be segmented accurately. This paper presents a new method for automatic segmentation of liver and tumor from CT images. The proposed method uses combination of intensity analysis, region-growing and preprocessing steps for automatic segmentation of liver and a second region growing process for tumor segmentation.

Children with optic pathway gliomas (OPGs), a low-grade brain tumor associated with neurofibromatosis type 1 (NF1-OPG), are at risk for permanent vision loss. While OPG size has been associated with vision loss, it is unclear how changes in size, shape, and imaging features of OPGs are associated with the likelihood of vision loss. This paper presents a fully automatic framework for accurate prediction of visual acuity loss using multi-sequence magnetic resonance images (MRIs). Our proposed framework includes a transformer-based segmentation network using transfer learning, statistical analysis of radiomic features, and a machine learning method for predicting vision loss. Our segmentation network was evaluated on multi-sequence MRIs acquired from 75 pediatric subjects with NF1-OPG and obtained an average Dice similarity coefficient of 0.791. The ability to predict vision loss was evaluated on a subset of 25 subjects with ground truth using cross-validation and achieved an average accuracy of 0.8. Analyzing multiple MRI features appear to be good indicators of vision loss, potentially permitting early treatment decisions.Clinical relevance- Accurately determining which children with NF1-OPGs are at risk and hence require preventive treatment before vision loss remains challenging, towards this we present a fully automatic deep learning-based framework for vision outcome prediction, potentially permitting early treatment decisions.

Breast cancer has become the primary factor threatening women's health. Automated breast volume scanner (ABVS) is applied for automatic scanning which is meaningful for the rapid and accurate detection of breast tumor. However, accurate segmentation of tumor regions is a huge challenge for clinicians from the ABVS images since it has the large image size and low data quality. Therefore, we propose a novel 3D deep convolutional neural network for automatic breast tumor segmentation from ABVS data. The structure based on 3D U-Net is designed with attention mechanism and transformer layers to optimize the extracted image features. In addition, we integrate the atrous spatial pyramid pooling block and the deep supervision for further performance improvement. The experimental results demonstrate that our model has achieved dice coefficient of 76.36% for 3D segmentation of breast tumor via our self-collected data.

To investigate the natural course of carotid plaque progression in transient ischemic attack/stroke patients by using serial multisequence magnetic resonance imaging (MRI). Forty transient ischemic attack/stroke patients with ipsilateral <70% carotid stenosis underwent MRI of the plaque ipsilateral to the symptomatic side at baseline and after 1 year. The MRI protocol consisted of T1-weighted turbo field-echo, time-of-flight, T2-weighted turbo spin-echo (TSE), and pre- and postgadopentetate dimeglumine-enhanced T1-weighted TSE images. For each plaque, carotid lumen volume, wall volume, total vessel volume (=carotid lumen volume + wall volume), the presence of a lipid-rich necrotic core (LRNC), fibrous cap (FC) status, and the presence of intraplaque hemorrhage (IPH) were assessed at both time points. Over a 1-year period, mean carotid lumen volume decreased with 4.8% ± 2.0% (±standard error) (P = 0.013). Mean wall volume increased with 11.2% ± 2.2% (P < 0.001). Total vessel volume did not significantly change (P = 0.147). At baseline, there were 18 plaques with a LRNC, which also had a LRNC at 1-year follow-up. No plaque without a LRNC at baseline developed a LRNC during the follow-up period. All plaques with a LRNC had a thin and/or ruptured FC at both time points. Twelve patients had IPH both at baseline and at follow-up. In one patient, IPH disappeared, whereas in another patient, new IPH appeared at follow-up. The presence of IPH and a LRNC with a thin and/or ruptured FC were not significantly associated with plaque progression (P > 0.05). In symptomatic patients with an ipsilateral carotid plaque causing <70% stenosis, we found evidence for inward plaque remodeling over a 1-year period. Overall, the presence/absence of IPH, a LRNC, and FC status did not change over 1 year.

This study aims to construct and validate a predictable deep learning model associated with clinical data and multi-sequence magnetic resonance imaging (MRI) for short-term postoperative facial nerve function in patients with acoustic neuroma. A total of 110 patients with acoustic neuroma who underwent surgery through the retrosigmoid sinus approach were included. Clinical data and raw features from four MRI sequences (T1-weighted, T2-weighted, T1-weighted contrast enhancement, and T2-weighted-Flair images) were analyzed. Spearman correlation analysis along with least absolute shrinkage and selection operator regression were used to screen combined clinical and radiomic features. Nomogram, machine learning, and convolutional neural network (CNN) models were constructed to predict the prognosis of facial nerve function on the seventh day after surgery. Receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were used to evaluate model performance. A total of 1050 radiomic parameters were extracted, from which 13 radiomic and 3 clinical features were selected. The CNN model performed best among all prediction models in the test set with an area under the curve (AUC) of 0.89 (95% CI, 0.84-0.91). CNN modeling that combines clinical and multi-sequence MRI radiomic features provides excellent performance for predicting short-term facial nerve function after surgery in patients with acoustic neuroma. As such, CNN modeling may serve as a potential decision-making tool for neurosurgery.

Herpes simplex virus of the brain (HSE) is a life-threatening neurological emergency which is characterized by high morbidity and mortality unless early diagnosed and treated. Recognition promptly is essential, because the mortality rate in untreated cases is about 70 percent whereas antiviral therapy decreases the death rate to 10-20 per cent in populations/patients treated and prevents long term neurological-cognitive sequelae. Developed magnetic resonance imaging (MRI) has been the key in the early detection of the HSE and the use of sequences like diffusion-weighted imaging (DWI), fluid-attenuated inversion recovery (FLAIR), and T2-weighted imaging has been found to be of high sensitivity and specificity. The lesions tend to be restricted to medial temporal lobes and limbic structures although extratemporal and bihemispheric damages have been reported. These imaging biomarkers can distinguish other pathologies, like neoplasm and cerebro vascular event, and can be employed to distinguish between HSE and autoimmune encephalitis (AE), which can both share similar clinical and radiologic phenotypes. The most recent advances in artificial intelligence (AI), which include classical machine learning (CML) and deep learning (DL) have shown significant potential to increase diagnostic accuracy. The presence of AI in combination with multi-sequence MRI allows automatic lesion detection, increased pattern recognition, and better classification between HSE, AE, and gliomas. Imaging with clinical and laboratory data such as cerebral things of analysis, serological marker and patient demographics have shown a better performance in diagnosis compared to the diagnostic performance of experienced radiologists and could predict functional outcome and cognitive recovery. However, a few issues restrict the use of multicenter imaging, such as inconsistent image schemes, small sample sizes, and technical demands of sophisticated imaging techniques like pseudo continuous arterial spin labeling (pCASL). Standardization of imaging workflows, larger multicenter and prospective validation studies must determine the reproducibility, generalization and clinical utility of imaging workflows. Synergistic use of high-tech MRI biomarkers with AI-related analytics is a perspective of early and differentiated diagnosis of HSE. These combined methods can help significantly improve patient outcomes and decrease the HSE-related burden of neurological morbidity because of timely antiviral intervention and individualized long-term care.

Nonlocal Means Denoising of Self-Gated and k-Space Sorted 4-Dimensional Magnetic Resonance Imaging Using Block-Matching and 3-Dimensional Filtering: Implications for Pancreatic Tumor Registration and Segmentation

We aimed to construct and validate a radiomics-based nomogram model derived from gadoxetic acid-enhanced magnetic resonance (MR) images to predict cytokeratin (CK) 19-positive (+) hepatocellular carcinoma (HCC) and patients' prognosis. A two-center and time-independent cohort of 311 patients were retrospectively enrolled (training cohort, n = 168; internal validation cohort, n = 72; external validation cohort, n = 71). A total of 2286 radiomic features were extracted from multisequence MR images with the uAI Research Portal (uRP), and a radiomic feature model was established. A combined model was established by incorporating the clinic-radiological features and the fusion radiomics signature using logistic regression analysis. Receiver operating characteristic curve (ROC) was used to evaluate the predictive efficacy of these models. Kaplan-Meier survival analysis was used to assess 1-year and 2-year progression-free survival (PFS) and overall survival (OS) in the cohort. By combining radiomic features extracted in DWI phase, arterial phase, venous and delay phase, the fusion radiomics signature achieved AUCs of 0.865, 0.824, and 0.781 in the training, internal, and external validation cohorts. The final combined clinic-radiological model showed higher AUC values in the three datasets compared with the fusion radiomics model. The nomogram based on the combined model showed satisfactory prediction performance in the training (C-index, 0.914), internal (C-index, 0.855), and external validation (C-index, 0.795) cohort. The 1-year and 2-year PFS and OS of the patients in the CK19+ group were 76% and 73%, and 78% and 68%, respectively. The 1-year and 2-year PFS and OS of the patients in the CK19-negative (-) group were 81% and 77%, and 80% and 74%, respectively. Kaplan-Meier survival analysis showed no significant differences in 1-year PFS and OS between the groups (P = 0.273 and 0.290), but it did show differences in 2-year PFS and OS between the groups (P = 0.032 and 0.040). Both PFS and OS were lower in CK19+ patients. The combined model based on clinic-radiological radiomics features can be used for predicting CK19+ HCC noninvasively to assist in the development of personalized treatment.

Attention-guided multi-scale learning network for automatic prostate and tumor segmentation on MRI

597 Background: Automated breast tumor identification and segmentation in magnetic resonance imaging (MRI) is a difficult and crucial area of study in breast cancer research. Artificial intelligence (AI) models are increasingly being developed for automated localization of lesions in imaging studies to facilitate quantitative assessment of features for improved diagnostic, prognostic and predictive performance. Such models have had success in detecting breast cancers in mammography, ultrasound and CT, but few have achieved three-dimensional (3D) volumetric tumor segmentation from breast MRI. The purpose of this study was to apply two state-of-the-art AI – specifically, deep learning (DL) - algorithms to 3D MRI breast cancer data and identify the higher performing algorithm for precise segmentation of breast tumors. Methods: We evaluated pre-treatment, T1 post-gadolinium contrast enhanced breast MRI from 222 patients with known breast cancers (n = 262). Images were split into training (n = 142), validation (n = 36), and hold-out test (n = 44) datasets. Two DL algorithms, U-Net and VAE-UNet, were trained to classify tumors on the training dataset across 1000 epochs. The output for each is a precise localization and segmentation of each tumor at the pixel level from every MRI image. We evaluated the performance of each algorithm using 5-fold cross-validation and testing on the validation and test sets. We calculated a dice accuracy score for each model as the performance comparison metric. Results: The highest dice accuracy score achieved on the validation dataset by generic U-Net was 83.38%, with an average across 1000 epochs of 62.41%. The highest dice accuracy achieved on the validation dataset by VAE-UNet was 82.62%, with an average across epochs of 61.28%. On our test dataset, the highest dice accuracy score achieved by U-Net was 93.09%, with an average across epochs of 66.31%, and the highest accuracy score for VAE-UNet was 90.98%, with average across epochs of 50.47%. Although U-Net appeared to perform slightly better than VAE-Unet for most cases, there were distinct cases where VAE-UNet outperformed U-Net (dice score up to 59% better than U-Net). Subsequent analysis indicated that VAE-UNet preferentially outperforms U-Net for tumors with low sphericity (p = 0.001). Conclusions: Our results suggest that U-Net is well suited for segmenting breast tumors from breast MRI in most cases, but that VAE-UNet outperforms U-Net when the tumor shapes are less spherical. Our findings could inform the choice of DL algorithms in research and clinical endeavors that rely on accurate breast cancer tumor segmentation. In particular, these two tools could be configured to facilitate tumor assessment from breast MRI in the clinical setting for: breast cancer screening in high-risk patient populations, pre-surgical planning, and monitoring of treatment response.

Nowadays, automatic tumor detection from brain images is extremely significant for many diagnostic as well as therapeutic purposes, due to the unpredictable shape and appearance of tumors. In medical image analysis, the automatic segmentation of tumors from brain using magnetic resonance imaging (MRI) data is the most critical issue. Existing research has some limitations, such as high processing time and lower accuracy, because of the time required for the training process. In this research, a new automatic segmentation process is introduced using machine learning and a swarm intelligence scheme. Here, a fuzzy logic with spiking neuron model (FL-SNM) is proposed for segmenting the brain tumor region in MR images. Initially, input images are preprocessed to remove Gaussian and Poisson noise using a modified Kuan filter (MKF). In the MKF, the optimal selection of the minimum MSE of image pixels is achieved using a random search algorithm (RSA), which improves the peak signal-to-noise ratio (PSNR). Then, the image is smoothed using an anisotropic diffusion filter (ADF) to reduce the over-filtering problem. Afterwards, to extract statistical texture features, Fisher's linear-discriminant analysis (FLDA) is used. Finally, extracted features are transferred to the FL-SNM process and this scheme effectively segments the tumor region. In FL-SNM, the consequent parameters such as weight and bias play an important role in segmenting the region. Therefore, optimizing the weight parameter values using a chicken behavior-based swarm intelligence (CSI) algorithm, is proposed. The proposed (FL-SNM) scheme attained better performance in terms of high accuracy (94.87%), sensitivity (92.07%), specificity (99.34%), precision rate (89.36%), recall rate (88.39%), F-measure (95.06%), G-mean (95.63%), and DSC rate (91.2%), compared to existing convolutional neural networks (CNNs) and hierarchical self-organizing maps (HSOMs).

Brain tumor segmentation in magnetic resonance imaging (MRI) is considered a complex procedure because of the variability of tumor shapes and the complexity of determining the tumor location, size, and texture. Manual tumor segmentation is a time-consuming task highly prone to human error. Hence, this study proposes an automated method that can identify tumor slices and segment the tumor across all image slices in volumetric MRI brain scans. First, a set of algorithms in the pre-processing stage is used to clean and standardize the collected data. A modified gray-level co-occurrence matrix and Analysis of Variance (ANOVA) are employed for feature extraction and feature selection, respectively. A multi-layer perceptron neural network is adopted as a classifier, and a bounding 3D-box-based genetic algorithm is used to identify the location of pathological tissues in the MRI slices. Finally, the 3D active contour without edge is applied to segment the brain tumors in volumetric MRI scans. The experimental dataset consists of 165 patient images collected from the MRI Unit of Al-Kadhimiya Teaching Hospital in Iraq. Results of the tumor segmentation achieved an accuracy of 89% ± 4.7% compared with manual processes.