MRI Segmentation Research Articles

Application of U-Net Network Utilizing Multiattention Gate for MRI Segmentation of Brain Tumors.

Studies have shown that the type of low-grade glioma is associated with its shape. The traditional diagnostic method involves extraction of the tumor shape from MRIs and diagnosing the type of glioma based on corresponding relationship between the glioma shape and type. This method is affected by the MRI background, tumor pixel size, and doctors' professional level, leading to misdiagnoses and missed diagnoses. With the help of deep learning algorithms, the shape of a glioma can be automatically segmented, thereby assisting doctors to focus more on the diagnosis of glioma and improving diagnostic efficiency. The segmentation of glioma MRIs using traditional deep learning algorithms exhibits limited accuracy, thereby impeding the effectiveness of assisting doctors in the diagnosis. The primary objective of our research is to facilitate the segmentation of low-grade glioma MRIs for medical practitioners through the utilization of deep learning algorithms. In this study, a UNet glioma segmentation network that incorporates multiattention gates was proposed to address this limitation. The UNet-based algorithm in the coding part integrated the attention gate into the hierarchical structure of the network to suppress the features of irrelevant regions and reduce the feature redundancy. In the decoding part, by adding attention gates in the fusion process of low- and high-level features, important feature information was highlighted, model parameters were reduced, and model sensitivity and accuracy were improved. The network model performed image segmentation on the glioma MRI dataset, and the accuracy and average intersection ratio (mIoU) of the algorithm segmentation reached 99.7%, 87.3%, 99.7%, and 87.6%. Compared with the UNet, PSPNet, and Attention UNet network models, this network model has obvious advantages in accuracy, mIoU, and loss convergence. It can serve as a standard for assisting doctors in diagnosis.

Automatic machine learning model for enhanced partition and identification of breast disorders in breast MRI scan

ABSTRACT The rapid identification and categorisation of breast cancers using low-contrast MRI images presents a significant challenge due to the disease’s prevalence among women of all ages. Nowadays, it is very difficult to classify and generalise models from low-contrast MRI datasets due to the prevalence of class imbalance caused by a wide range of symptoms and untrustworthy data sources. Using ensemble learning with optimised k-means helps with these problems; it reduces overfitting and improves reliability by using methods like k-means clustering, frog feap algorithm (FLA), boosting, and bagging to balance datasets and show the complexity of symptoms. In this research, we present a new strategy that makes use of a hybrid optimised K-Means algorithm combined with the FLA, along with thresholding and morphological approaches, and ensemble learning. There are two main stages to our work. In the initial phase, we preprocess low-contrast MRI images and delineate the breast tumour area from the segmented MRI images. Furthermore, we employ Discrete Wavelet Transformation (DWT) to extract the most salient features. In the next step, the retrieved features will be used as input parameters for an ensemble-learned breast tumour classifier. To determine whether a low-contrast breast tumour MRI picture has a benign or malignant tumour after model training, we use a variety of classifiers, such as k-nearest neighbours (KNN), decision trees (DT), gradient boosting (GB), random forests (RF), and artificial neural networks (ANN). Through a voting mechanism that averages the accuracy of all models, our suggested framework achieves a remarkable 98.8% accuracy rate. In addition, proposed model is quite efficient & reliable to detect the kind of breast tumour or cancer with a sensitivity of 98.02% and a specificity of 97.5%.

Assessing Aortic Motion with Automated 3D Cine Balanced SSFP MRI Segmentation

PurposeTo apply free-running three-dimensional (3D) cine balanced steady state free precession (bSSFP) CMR framework in combination with AI segmentations to quantify time-resolved aortic displacement, diameter and diameter change. MethodsIn this prospective study, we implemented a free-running 3D cine bSSFP sequence with scan time of about 4minutes facilitated by pseudo-spiral Cartesian undersampling and compressed-sensing reconstruction. Automated segmentation of all cardiac timeframes was applied through the use of nnU-Net. Dynamic 3D motion maps were created for three repeated scans per volunteer, leading to the detailed quantification of motion, as well as the measurement and change in diameter of the ascending aorta. ResultsA total of 14 adult healthy volunteers (median age, 28 years (IQR: 26.0-31.3), 6 female) were included. Automated segmentation compared to manual segmentation of the aorta test set showed a Dice score of 0.93 ± 0.02. The median (interquartile range) over all volunteers for the largest maximum and mean ascending aorta (AAo) displacement in the first scan was 13.0 (4.4) mm and 5.6 (2.4) mm, respectively. Peak mean diameter in the AAo was 25.9 (2.2) mm and peak mean diameter change was 1.4 (0.5) mm. The maximum individual variability over the three repeated scans of maximum and mean AAo displacement was 3.9 (1.6) mm and 2.2 (0.8) mm, respectively. The maximum individual variability of mean diameter and diameter change were 1.2 (0.5) mm and 0.9 (0.4) mm. ConclusionA free-running 3D cine bSSFP CMR scan with a scan time of four minutes combined with an automated nnU-net segmentation consistently captured the aorta’s cardiac motion-related 4D displacement, diameter, and diameter change.

Development of an initial training and evaluation programme for manual lower limb muscle MRI segmentation

BackgroundMagnetic resonance imaging (MRI) quantification of intramuscular fat accumulation is a responsive biomarker in neuromuscular diseases. Despite emergence of automated methods, manual muscle segmentation remains an essential foundation. We aimed to develop a training programme for new observers to demonstrate competence in lower limb muscle segmentation and establish reliability benchmarks for future human observers and machine learning segmentation packages.MethodsThe learning phase of the training programme comprised a training manual, direct instruction, and eight lower limb MRI scans with reference standard large and small regions of interest (ROIs). The assessment phase used test–retest scans from two patients and two healthy controls. Interscan and interobserver reliability metrics were calculated to identify underperforming outliers and to determine competency benchmarks.ResultsThree experienced observers undertook the assessment phase, whilst eight new observers completed the full training programme. Two of the new observers were identified as underperforming outliers, relating to variation in size or consistency of segmentations; six had interscan and interobserver reliability equivalent to those of experienced observers. The calculated benchmark for the Sørensen-Dice similarity coefficient between observers was greater than 0.87 and 0.92 for individual thigh and calf muscles, respectively. Interscan and interobserver reliability were significantly higher for large than small ROIs (all p < 0.001).ConclusionsWe developed, implemented, and analysed the first formal training programme for manual lower limb muscle segmentation. Large ROI showed superior reliability to small ROI for fat fraction assessment.Relevance statementObservers competent in lower limb muscle segmentation are critical to application of quantitative muscle MRI biomarkers in neuromuscular diseases. This study has established competency benchmarks for future human observers or automated segmentation methods.Key points• Observers competent in muscle segmentation are critical for quantitative muscle MRI biomarkers.• A training programme for muscle segmentation was undertaken by eight new observers.• We established competency benchmarks for future human observers or automated segmentation methods.Graphical

Attention-Based Neural Network for Cardiac MRI Segmentation: Application to Strain and Volume Computation

ContextDeep learning algorithms have been widely used for cardiac image segmentation. However, most of these architectures rely on convolutions that hardly model long-range dependencies, limiting their ability to extract contextual information. Moreover, the traditional U-net architecture suffers from the difference of semantic information between feature maps of the encoder and decoder (also known as the semantic gap). Material and methodTo address this issue, a new network architecture relying on attention mechanism was introduced. Swin Filtering Blocks (SFB), that use Swin Transformer blocks in a cross-attention manner, were added between the encoder and the decoder to filter information coming from the encoder based on the feature map from the decoder. Attention was also employed at the lowest resolution in the form of a transformer layer to increase the receptive field of the network.We conducted experiments to assess both generalization capability and to evaluate how training on all frames of the cardiac cycle rather than only the end-diastole and end-systole impacts strain and segmentation performances. Results and conclusionVisual inspection of feature maps suggested that Swin Filtering Blocks contribute to the reduction of the semantic gap. Performing attention between all patches using a transformer layer brought higher performance than convolutions. Training the model with all phases of the cardiac cycle resulted in slightly more accurate segmentations while leading to a more noticeable improvement for strain estimation. A limited decrease in performance was observed when testing on out-of-distribution data, but the gap widens for the most apical slices.

Fully automated segmentation and volumetric measurement of ocular adnexal lymphoma by deep learning-based self-configuring nnU-net on multi-sequence MRI: a multi-center study

PurposeTo evaluate nnU-net’s performance in automatically segmenting and volumetrically measuring ocular adnexal lymphoma (OAL) on multi-sequence MRI.MethodsWe collected T1-weighted (T1), T2-weighted and T1-weighted contrast-enhanced images with/without fat saturation (T2_FS/T2_nFS, T1c_FS/T1c_nFS) of OAL from four institutions. Two radiologists manually annotated lesions as the ground truth using ITK-SNAP. A deep learning framework, nnU-net, was developed and trained using two models. Model 1 was trained on T1, T2, and T1c, while Model 2 was trained exclusively on T1 and T2. A 5-fold cross-validation was utilized in the training process. Segmentation performance was evaluated using the Dice similarity coefficient (DSC), sensitivity, and positive prediction value (PPV). Volumetric assessment was performed using Bland-Altman plots and Lin’s concordance correlation coefficient (CCC).ResultsA total of 147 patients from one center were selected as training set and 33 patients from three centers were regarded as test set. For both Model 1 and 2, nnU-net demonstrated outstanding segmentation performance on T2_FS with DSC of 0.80–0.82, PPV of 84.5–86.1%, and sensitivity of 77.6–81.2%, respectively. Model 2 failed to detect 19 cases of T1c, whereas the DSC, PPV, and sensitivity for T1_nFS were 0.59, 91.2%, and 51.4%, respectively. Bland–Altman plots revealed minor tumor volume differences with 0.22–1.24 cm3 between nnU-net prediction and ground truth on T2_FS. The CCC were 0.96 and 0.93 in Model 1 and 2 for T2_FS images, respectively.ConclusionThe nnU-net offered excellent performance in automated segmentation and volumetric assessment in MRI of OAL, particularly on T2_FS images.

Stepwise Transfer Learning for Expert-level Pediatric Brain Tumor MRI Segmentation in a Limited Data Scenario.

Purpose To develop, externally test, and evaluate clinical acceptability of a deep learning pediatric brain tumor segmentation model using stepwise transfer learning. Materials and Methods In this retrospective study, the authors leveraged two T2-weighted MRI datasets (May 2001 through December 2015) from a national brain tumor consortium (n = 184; median age, 7 years [range, 1-23 years]; 94 male patients) and a pediatric cancer center (n = 100; median age, 8 years [range, 1-19 years]; 47 male patients) to develop and evaluate deep learning neural networks for pediatric low-grade glioma segmentation using a stepwise transfer learning approach to maximize performance in a limited data scenario. The best model was externally tested on an independent test set and subjected to randomized blinded evaluation by three clinicians, wherein they assessed clinical acceptability of expert- and artificial intelligence (AI)-generated segmentations via 10-point Likert scales and Turing tests. Results The best AI model used in-domain stepwise transfer learning (median Dice score coefficient, 0.88 [IQR, 0.72-0.91] vs 0.812 [IQR, 0.56-0.89] for baseline model; P = .049). With external testing, the AI model yielded excellent accuracy using reference standards from three clinical experts (median Dice similarity coefficients: expert 1, 0.83 [IQR, 0.75-0.90]; expert 2, 0.81 [IQR, 0.70-0.89]; expert 3, 0.81 [IQR, 0.68-0.88]; mean accuracy, 0.82). For clinical benchmarking (n = 100 scans), experts rated AI-based segmentations higher on average compared with other experts (median Likert score, 9 [IQR, 7-9] vs 7 [IQR 7-9]) and rated more AI segmentations as clinically acceptable (80.2% vs 65.4%). Experts correctly predicted the origin of AI segmentations in an average of 26.0% of cases. Conclusion Stepwise transfer learning enabled expert-level automated pediatric brain tumor autosegmentation and volumetric measurement with a high level of clinical acceptability. Keywords: Stepwise Transfer Learning, Pediatric Brain Tumors, MRI Segmentation, Deep Learning Supplemental material is available for this article. © RSNA, 2024.

3D Observation of Pelvic Organs with Dynamic MRI Segmentation: A Bridge Toward Patient-Specific Models.

Female pelvic organ prolapses are common, but their treatment is challenging. Notably, diagnosis and understanding of these troubles remain incomplete. Tridimensional observations of displacement and deformation of the pelvic organs during a strain could support a better understanding and help to develop comprehensive tools for preoperative planning. The present feasibility study evaluates tridimensional dynamic MRI in 12 healthy volunteers. Tridimensional acquisitions were approximated using five intersecting slices, each recorded twice per second. MRI was performed during rest and strain, with intrarectal and intravaginal contrast gel. Subject-specific dynamic 3D models were built for each volunteer through segmentation. For each volunteer, pelvic organs could be segmented in three dimensions with a rate of acquisition of two cycles per second on five slices, allowing for a fluid observation of displacements and deformations during strain. Manual segmentation of a full strain required 2 h and 33 min on average. The upper limit of the rectum and the pelvic floor were the most difficult structures to identify. This technique is limited by its time-consuming manual segmentation, which impedes its implantation for routine clinical use. This method must be tried in patients with pelvic organ prolapse. This multi-planar acquisition technique applied during a dynamic MRI allows for observation of displacement and deformations of pelvic organs during a strain.

RobU-Net: a heuristic robust multi-class brain tumor segmentation approaches for MRI scans

A tumor is an abnormal growth of cells, either cancerous or benign, that develops in an organ. Early detection and segmentation of brain tumors are crucial for effective treatment, but manual analysis by experts is a labor-intensive and time-consuming process. The proposed solution is a deep learning framework called RobU-Net, a modified U-Net, to improve the accuracy and robustness of MRI image segmentation, specifically designed to handle Rician noise in MRI scans. The study uses various encoder-decoder architectures to segment MRI scans, achieving optimal results that match the ground truth. The approach also employs discrete wavelet transforms for contrast enhancement and tests the framework's robustness. This study used the MRI brain tumor dataset with 3064 slices. The space and time analysis have been carried out using sensitivity (Sn), dice loss (DL), dice coefficient (DC), and Jaccard index (JI). A comprehensive assessment of our proposed framework depicted that it is better than recently existing frameworks. We have found the so far highest Sn, DC, and JI values for noise-impregnated and original MRI datasets using two novice architectures (overall best for A1-E2 and optimum or noisy dataset for A2-E1) as (0.9831, 0.9726, and 0.9468) and (0.9199, 0.9781, and 0.9571).

78-PUB: Machine Learning–Assisted Segmentation of Pancreas MRI

78-PUB: Machine Learning–Assisted Segmentation of Pancreas MRI

Integration of pre-treatment computational radiomics, deep radiomics, and transcriptomics enhances soft-tissue sarcoma patient prognosis

Our objective was to capture subgroups of soft-tissue sarcoma (STS) using handcraft and deep radiomics approaches to understand their relationship with histopathology, gene-expression profiles, and metastatic relapse-free survival (MFS). We included all consecutive adults with newly diagnosed locally advanced STS (N = 225, 120 men, median age: 62 years) managed at our sarcoma reference center between 2008 and 2020, with contrast-enhanced baseline MRI. After MRI postprocessing, segmentation, and reproducibility assessment, 175 handcrafted radiomics features (h-RFs) were calculated. Convolutional autoencoder neural network (CAE) and half-supervised CAE (HSCAE) were trained in repeated cross-validation on representative contrast-enhanced slices to extract 1024 deep radiomics features (d-RFs). Gene-expression levels were calculated following RNA sequencing (RNAseq) of 110 untreated samples from the same cohort. Unsupervised classifications based on h-RFs, CAE, HSCAE, and RNAseq were built. The h-RFs, CAE, and HSCAE grouping were not associated with the transcriptomics groups but with prognostic radiological features known to correlate with lower survivals and higher grade and SARCULATOR groups (a validated prognostic clinical-histological nomogram). HSCAE and h-RF groups were also associated with MFS in multivariable Cox regressions. Combining HSCAE and transcriptomics groups significantly improved the prognostic performances compared to each group alone, according to the concordance index. The combined radiomic-transcriptomic group with worse MFS was characterized by the up-regulation of 707 genes and 292 genesets related to inflammation, hypoxia, apoptosis, and cell differentiation. Overall, subgroups of STS identified on pre-treatment MRI using handcrafted and deep radiomics were associated with meaningful clinical, histological, and radiological characteristics, and could strengthen the prognostic value of transcriptomics signatures.

Evaluating Sex-specific Differences in Abdominal Fat Volume and Proton Density Fat Fraction at MRI Using Automated nnU-Net-based Segmentation.

Sex-specific abdominal organ volume and proton density fat fraction (PDFF) in people with obesity during a weight loss intervention was assessed with automated multiorgan segmentation of quantitative water-fat MRI. An nnU-Net architecture was employed for automatic segmentation of abdominal organs, including visceral and subcutaneous adipose tissue, liver, and psoas and erector spinae muscle, based on quantitative chemical shift-encoded MRI and using ground truth labels generated from participants of the Lifestyle Intervention (LION) study. Each organ's volume and fat content were examined in 127 participants (73 female and 54 male participants; body mass index, 30-39.9 kg/m2) and in 81 (54 female and 32 male participants) of these participants after an 8-week formula-based low-calorie diet. Dice scores ranging from 0.91 to 0.97 were achieved for the automatic segmentation. PDFF was found to be lower in visceral adipose tissue compared with subcutaneous adipose tissue in both male and female participants. Before intervention, female participants exhibited higher PDFF in subcutaneous adipose tissue (90.6% vs 89.7%; P < .001) and lower PDFF in liver (8.6% vs 13.3%; P < .001) and visceral adipose tissue (76.4% vs 81.3%; P < .001) compared with male participants. This relation persisted after intervention. As a response to caloric restriction, male participants lost significantly more visceral adipose tissue volume (1.76 L vs 0.91 L; P < .001) and showed a higher decrease in subcutaneous adipose tissue PDFF (2.7% vs 1.5%; P < .001) than female participants. Automated body composition analysis on quantitative water-fat MRI data provides new insights for understanding sex-specific metabolic response to caloric restriction and weight loss in people with obesity. Keywords: Obesity, Chemical Shift-encoded MRI, Abdominal Fat Volume, Proton Density Fat Fraction, nnU-Net ClinicalTrials.gov registration no. NCT04023942 Supplemental material is available for this article. Published under a CC BY 4.0 license.

Automatic segmentation of medial temporal lobe subregions in multi-scanner, multi-modality MRI of variable quality.

Volumetry of subregions in the medial temporal lobe (MTL) computed from automatic segmentation in MRI can track neurodegeneration in Alzheimer's disease. However, image quality may vary in MRI. Poor quality MR images can lead to unreliable segmentation of MTL subregions. Considering that different MRI contrast mechanisms and field strengths (jointly referred to as "modalities" here) offer distinct advantages in imaging different parts of the MTL, we developed a muti-modality segmentation model using both 7 tesla (7T) and 3 tesla (3T) structural MRI to obtain robust segmentation in poor-quality images. MRI modalities including 3T T1-weighted, 3T T2-weighted, 7T T1-weighted and 7T T2-weighted (7T-T2w) of 197 participants were collected from a longitudinal aging study at the Penn Alzheimer's Disease Research Center. Among them, 7T-T2w was used as the primary modality, and all other modalities were rigidly registered to the 7T-T2w. A model derived from nnU-Net took these registered modalities as input and outputted subregion segmentation in 7T-T2w space. 7T-T2w images most of which had high quality from 25 selected training participants were manually segmented to train the multi-modality model. Modality augmentation, which randomly replaced certain modalities with Gaussian noise, was applied during training to guide the model to extract information from all modalities. To compare our proposed model with a baseline single-modality model in the full dataset with mixed high/poor image quality, we evaluated the ability of derived volume/thickness measures to discriminate Amyloid+ mild cognitive impairment (A+MCI) and Amyloid- cognitively unimpaired (A-CU) groups, as well as the stability of these measurements in longitudinal data. The multi-modality model delivered good performance regardless of 7T-T2w quality, while the single-modality model under-segmented subregions in poor-quality images. The multi-modality model generally demonstrated stronger discrimination of A+MCI versus A-CU. Intra-class correlation and Bland-Altman plots demonstrate that the multi-modality model had higher longitudinal segmentation consistency in all subregions while the single-modality model had low consistency in poor-quality images. The multi-modality MRI segmentation model provides an improved biomarker for neurodegeneration in the MTL that is robust to image quality. It also provides a framework for other studies which may benefit from multimodal imaging.

Deep learning-based liver cyst segmentation in MRI for autosomal dominant polycystic kidney disease

Abstract Background Autosomal dominant polycystic kidney disease (ADPKD) can lead to polycystic liver disease (PLD), characterized by liver cysts. Although majority of the patients are asymptomatic, massively enlarged liver secondary to PLD can cause discomfort, and compression on adjacent structures requiring cyst aspiration/fenestration, partial liver resection, or liver transplantation. Monitoring PLD by measuring liver volume fails to track the early stages when liver cyst volume is too small to affect liver volume. Purpose To improve PLD assessment in the early stages by automating detection and segmentation of liver cysts using deep learning (DL) models. Materials and Methods A self-configured UNet-based platform (nnU-Net) was trained with 40 ADPKD subjects with liver cysts annotated by a radiologist. Internal (n = 7), External (n = 10), and test-retest reproducibility (n = 17) validations included macro- and micro-level performance metrics: patient-level Dice scores (PDice), along with voxel-level true positive rates (VTPR), as well as analysis of time saved in a model-assisted scenario. Additionally, we assessed human-level reliability in liver cyst segmentation and evaluated the model’s test-retest reproducibility. We further compared liver volume vs cyst volume for tracking disease in a subject with 16+ years follow-up. Results The model achieved an 82% ± 11% PDice and a 75% ± 15% VTPR on the internal test sets (n = 7 patients), and 80% ± 12% Dice score and a 91% ± 7% VTPR on the external test sets (n = 10 patients). It excelled particularly in detecting small liver cysts, a challenging task for manual annotation. This efficiency translated to a median of 91% (IQR: 14%) reduction in annotation time compared to manual labeling. Test-retest assessment demonstrated excellent reproducibility, with coefficients of variation of 94% for liver cyst fraction and 92% for cyst count. Conclusion DL automation of liver cyst segmentations demonstrates potential to improve tracking of liver cyst volume in polycystic liver disease.

Deep Learning for Nasopharyngeal Carcinoma Segmentation in Magnetic Resonance Imaging: A Systematic Review and Meta-Analysis

Nasopharyngeal carcinoma is a significant health challenge that is particularly prevalent in Southeast Asia and North Africa. MRI is the preferred diagnostic tool for NPC due to its superior soft tissue contrast. The accurate segmentation of NPC in MRI is crucial for effective treatment planning and prognosis. We conducted a search across PubMed, Embase, and Web of Science from inception up to 20 March 2024, adhering to the PRISMA 2020 guidelines. Eligibility criteria focused on studies utilizing DL for NPC segmentation in adults via MRI. Data extraction and meta-analysis were conducted to evaluate the performance of DL models, primarily measured by Dice scores. We assessed methodological quality using the CLAIM and QUADAS-2 tools, and statistical analysis was performed using random effects models. The analysis incorporated 17 studies, demonstrating a pooled Dice score of 78% for DL models (95% confidence interval: 74% to 83%), indicating a moderate to high segmentation accuracy by DL models. Significant heterogeneity and publication bias were observed among the included studies. Our findings reveal that DL models, particularly convolutional neural networks, offer moderately accurate NPC segmentation in MRI. This advancement holds the potential for enhancing NPC management, necessitating further research toward integration into clinical practice.

Multimodal MRI segmentation of key structures for microvascular decompression via knowledge-driven mutual distillation and topological constraints.

Microvascular decompression (MVD) is a widely used neurosurgical intervention for the treatment of cranial nerves compression. Segmentation of MVD-related structures, including the brainstem, nerves, arteries, and veins, is critical for preoperative planning and intraoperative decision-making. Automatically segmenting structures related to MVD is still challenging for current methods due to the limited information from a single modality and the complex topology of vessels and nerves. Considering that it is hard to distinguish MVD-related structures, especially for nerve and vessels with similar topology, we design a multimodal segmentation network with a shared encoder-dual decoder structure and propose a clinical knowledge-driven distillation scheme, allowing reliable knowledge transferred from each decoder to the other. Besides, we introduce a class-wise contrastive module to learn the discriminative representations by maximizing the distance among classes across modalities. Then, a projected topological loss based on persistent homology is proposed to constrain topological continuity. We evaluate the performance of our method on in-house dataset consisting of 100 paired HR-T2WI and 3D TOF-MRA volumes. Experiments indicate that our model outperforms the SOTA in DSC by 1.9% for artery, 3.3% for vein and 0.5% for nerve. Visualization results show our method attains improved continuity and less breakage, which is also consistent with intraoperative images. Our method can comprehensively extract the distinct features from multimodal data to segment the MVD-related key structures and preserve the topological continuity, allowing surgeons precisely perceiving the patient-specific target anatomy and substantially reducing the workload of surgeons in the preoperative planning stage. Our resources will be publicly available at https://github.com/JaronTu/Multimodal_MVD_Seg .

Dynamic Regularized Adaptive Cluster Optimization (DRACO) for Quantitative Cardiac Cine MRI in Complex Arrhythmias.

Irregular cardiac motion can render conventional segmented cine MRI nondiagnostic. Clustering has been proposed for cardiac motion binning and may be optimized for complex arrhythmias. To develop an adaptive cluster optimization method for irregular cardiac motion, and to generate the corresponding time-resolved cine images. Prospective. Thirteen with atrial fibrillation, four with premature ventricular contractions, and one patient in sinus rhythm. Free-running balanced steady state free precession (bSSFP) with sorted golden-step, reference real-time sequence. Each subject underwent both the sorted golden-step bSSFP and the reference Cartesian real-time imaging. Golden-step bSSFP images were reconstructed using the dynamic regularized adaptive cluster optimization (DRACO) method and k-means clustering. Image quality (4-point Likert scale), signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), edge sharpness, and ventricular function were assessed. Paired t-tests, Friedman test, regression analysis, Fleiss' Kappa, Bland-Altman analysis. Significance level P < 0.05. The DRACO method had the highest percent of images with scores ≥3 (96% for diastolic frame, 93% for systolic frame, and 93% for multiphase cine) and the percentages were significantly higher compared with both the k-means and real-time methods. Image quality scores, SNR, and CNR were significantly different between DRACO vs. k-means and between DRACO vs. real-time. Cardiac function analysis showed no significant differences between DRACO vs. the reference real-time. DRACO with time-resolved reconstruction generated high quality images and has early promise for quantitative cine cardiac MRI in patients with complex arrhythmias including atrial fibrillation. Stage 2.

VINNA for Neonates – Orientation Independence through Latent Augmentations

Abstract A robust, fast, and accurate segmentation of neonatal brain images is highly desired to better understand and detect changes during development and disease, specifically considering the rise in imaging studies for this cohort. Yet, the limited availability of ground truth datasets, lack of standardized acquisition protocols, and wide variations of head positioning in the scanner pose challenges for method development. A few automated image analysis pipelines exist for newborn brain MRI segmentation, but they often rely on time-consuming non-linear spatial registration procedures and require resampling to a common resolution, subject to loss of information due to interpolation and down-sampling. Without registration and image resampling, variations with respect to head positions and voxel resolutions have to be addressed differently. In deep-learning, external augmentations such as rotation, translation, and scaling are traditionally used to artificially expand the representation of spatial variability, which subsequently increases both the training dataset size and robustness. However, these transformations in the image space still require resampling, reducing accuracy specifically in the context of label interpolation. We recently introduced the concept of resolution-independence with the Voxel-size Independent Neural Network framework, VINN. Here, we extend this concept by additionally shifting all rigid-transforms into the network architecture with a four degree of freedom (4-DOF) transform module, enabling resolution-aware internal augmentations (VINNA) for deep learning. In this work we show that VINNA (i) significantly outperforms state-of-the-art external augmentation approaches, (ii) effectively addresses the head variations present specifically in newborn datasets, and (iii) retains high segmentation accuracy across a range of resolutions (0.5-1.0mm). Furthermore, the 4-DOF transform module together with internal augmentations is a powerful, general approach to implement spatial augmentation without requiring image or label interpolation. The specific network application to newborns will be made publicly available as VINNA4neonates.

Abstract PO2-02-11: Radiomics and an Erosion Model to Predict Post-Neoadjuvant Therapy Tumor Characteristics for Personalized Breast Cancer Treatment

Abstract Background: Neoadjuvant therapy (NAT) has become a standard-of-care (SOC) for patients diagnosed with locally advanced or early-stage breast cancer. A critical step towards optimizing treatment selection and coordinating surgical care is to estimate the tumor’s response to NAT. While pathological complete response (pCR) is commonly used to measure NAT effectiveness, patients who do not achieve pCR still benefit from NAT depending on the tumor’s response. A reduction in tumor extent could alter the surgical approach from mastectomy to breast conserving therapy and improve recurrence and overall survival. In this study we used pre-treatment imaging data to provide a comprehensive prediction of clinically relevant metrics including tumor volume, tumor convex hull volume, and spatially derived metrics; this research represents an important advancement in predicting NAT effectiveness. Methods: 262 breast cancer patients from 5 institutions who underwent NAT with pre- and post-NAT dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) were included in this study. Each MRI was automatically segmented using a convolutional neural network with manual edits and verification. Radiomic features were extracted from the pre-NAT MRI and tumor segmentation. Significant radiomic features were selected based on their correlation with post-NAT volume, feature-to-feature correlation, and LASSO regression. Altogether, 12 and 17 radiomic features were selected for volume and convex hull volume prediction, respectively. The selected features were inputted into a Huber loss regression model to predict either the post-NAT tumor volume or the convex hull volume (surgical excision volume). To enable spatial and morphological comparison between the predicted and post-NAT tumor, an erosion model was developed in which the pre-NAT and post-NAT MRIs were co-registered, and the pre-NAT tumor was symmetrically eroded until its volume matched the predicted volume. Clinically relevant metrics such as the distance from the tumor to the nipple, skin, and chest were computed to assess the morphological and spatial prediction capabilities of the model. All models were tested on patients from cohort holdout sets (n=110) unseen during training. Results: The volume prediction analysis yielded promising results, with an R^2 value of 0.62 for post-NAT tumor volume prediction and an R^2 value of 0.56 for post-NAT tumor convex hull volume prediction. The mean volume and convex hull volume error was 1.1 cc and 1.3 cc, respectively. Table 1 provides an overview of the morphological and spatial comparisons between the predicted and post-NAT tumors. Overall, the predicted tumors demonstrated close agreement with their post-NAT counterparts. Conclusion: This study demonstrates a successful approach towards estimating post-NAT tumor volume, convex hull volume, and morphological and spatial characteristics. By providing reliable estimates of post-NAT tumor characteristics using SOC pre-NAT data, our predictive models enable personalized treatment planning and patient stratification to optimize patient care. Table 1. Predicted vs. post-NAT tumor comparison. Results demonstrate accurate predictions for clinically relevant distance metrics. Citation Format: Bradley Feiger, John Pfeiffer, John Cole, Anuja Antony, Joseph Peterson. Radiomics and an Erosion Model to Predict Post-Neoadjuvant Therapy Tumor Characteristics for Personalized Breast Cancer Treatment [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO2-02-11.

ConvNextUNet: A small-region attentioned model for cardiac MRI segmentation

Cardiac MRI segmentation is a significant research area in medical image processing, holding immense clinical and scientific importance in assisting the diagnosis and treatment of heart diseases. Currently, existing cardiac MRI segmentation algorithms are often constrained by specific datasets and conditions, leading to a notable decrease in segmentation performance when applied to diverse datasets. These limitations affect the algorithm’s overall performance and generalization capabilities. Inspired by ConvNext, we introduce a two-dimensional cardiac MRI segmentation U-shaped network called ConvNextUNet. It is the first application of a combination of ConvNext and the U-shaped architecture in the field of cardiac MRI segmentation. Firstly, we incorporate up-sampling modules into the original ConvNext architecture and combine it with the U-shaped framework to achieve accurate reconstruction. Secondly, we integrate Input Stem into ConvNext, and introduce attention mechanisms along the bridging path. By merging features extracted from both the encoder and decoder, a probability distribution is obtained through linear and nonlinear transformations, serving as attention weights, thereby enhancing the signal of the same region of interest. The resulting attention weights are applied to the decoder features, highlighting the region of interest. This allows the model to simultaneously consider local context and global details during the learning phase, fully leveraging the advantages of both global and local perception for a more comprehensive understanding of cardiac anatomical structures. Consequently, the model demonstrates a clear advantage and robust generalization capability, especially in small-region segmentation. Experimental results on the ACDC, LVQuan19, and RVSC datasets confirm that the ConvNextUNet model outperforms the current state-of-the-art models, particularly in small-region segmentation tasks. Furthermore, we conducted cross-dataset training and testing experiments, which revealed that the pre-trained model can accurately segment diverse cardiac datasets, showcasing its powerful generalization capabilities. The source code of this project is available at https://github.com/Zemin-Cai/ConvNextUNet.