Keywords: AI/ML Software, Machine Learning/Artificial Intelligence, Active Learning, Tractography, White Matter

Motivation: Accurate tractography-based segmentation of white matter tracts is crucial for tasks such as pre-surgical planning. Fully automated methods are limited to predefined tracts and struggle with anatomical deviations, e.g. caused by tumors.

Goal(s): Our goal is to enhance the manual segmentation process through a novel and intuitive approach.

Approach: We recently developed atTRACTive, a tool for semi-automatic fiber dissection relying on entropy-based active learning. In this work, we have improved atTRACTive and conducted an initial evaluation of its test-retest reliability in comparison to traditional ROI-based tract segmentation methods.

Results: atTRACTive has demonstrated superior test-retest reliability compared to traditional ROI-based segmentation approaches.

Impact: The method offers guidance to researchers in the intuitive and efficient segmentation of arbitrary white matter tracts. Instead of drawing challenging-to-reproduce ROIs, users can simply annotate meaningful streamlines, which are then used to train a classifier.

Keywords: AI/ML Software, Machine Learning/Artificial Intelligence, Specific absorption rate (SAR), artifacts, ultra-high field MRI, deep learning

Motivation: Gridding artifacts in neural network estimated images are common and could inhibit neural network estimation accuracy.

Goal(s): The goal is to discover which parameters are responsible for gridding artifacts, and whether the artifacts inhibit estimation quality. 

Approach:  We test the effects of simulated body models, neural network parameters, and postprocessing methods on gridding artifacts, and their effect on overall neural network estimation accuracy in the context of local specific energy absorption rate (SAR) matrices, a patient safety concern for MRI scanning. 

Results: Altering neural network parameters affects the presentation of gridding artifacts the most. Eliminating gridding artifacts improves network estimation accuracy. 

Impact: Researchers working with computer vision whose images experience a gridding artifact can inform their neural network parameter tuning efforts with the results of this exploratory study.  

Keywords: AI/ML Software, Cancer

Motivation: To enhance nasopharyngeal carcinoma (NPC) diagnostics, this study aims to assess the accuracy and image quality of relaxometry maps using fast synthetic MRI with deep learning reconstruction.

Goal(s): The primary goal is to evaluate the potential of DL Recon for NPC diagnosis, focusing on reduce scan time, improve image quality and quantitative accuracy to enable early lesion detection.

Approach: Two protocols (Trad: lower acceleration rate without DL Recon, DLR: higher acceleration rate with DL Recon) was performed on twenty-four NPC patients to evaluated T1/T2/PD measurements and image quality.

Results: Fast MAGiC acquisition with DL Recon can retain accuracy and improve image quality.

Impact: With DL Recon, the MAGiC acquisition can achieve in shorter scan time, with enhanced image quality and maintained quantitative accuracy in NPC diagnosis use.

Keywords: AI Diffusion Models, Pelvis

Motivation: Mathematical, biophysical, and/or statistical models are typically used to analyze diffusion-weighted imaging signals, yielding quantitative biomarkers. Those model-based approaches, however, often suffer from limited model capability, fitting instability, and degeneracy. 

Goal(s): To use a MOdel-free Diffusion-wEighted MRI technique (MODEM) to differentiate underlying tissues based on diffusion signal intensities. 

Approach: We developed a machine-learning-based approach which we call MOdel-free Diffusion-wEighted MRI technique(MODEM) and assess its performance by using synthetic DWI data from Monte Carlo simulations and cervical staging dataset.

Results: MODEM exhibited superior diagnostic performance to the model-based approach in both Monte Carlo simulations and cervical cancer staging data.

Impact: A model-free machine-learning-based approach provides superior performance to the conventional diffusion-model-based approach for differentiating the underlying tissue properties. 

Keywords: Head & Neck/ENT, Machine Learning/Artificial Intelligence

Motivation: Preoperative distinguishment of the benign parotid gland tumors from the malignant determines the surgical scope; however, identifying tumor nature with only T1-weighted and fat-suppressed T2-weighted images is challenging.

Goal(s): Inserting three adjacent slices of the tumor into the RGB channels of a 2D image as 2.5D images coupled with transfer learning was utilized.

Approach: Using 2D and 2.5D images as input, a ResNet-101 model, pre-trained on ImageNet, was employed for transfer learning to facilitate the prediction.

Results: Deep learning models discerned malignant parotid gland tumors from the benign, especially 2.5D model showed superior performance to 2D model.

Impact: The transfer learning and 2.5D-MRI based classification model offered new insights to differentiate the malignant parotid gland tumors from the benign ones, especially when sample quantities are limited.

Keywords: Other AI/ML, Segmentation, Supervised Deep Learning; Data Labeling

Motivation: Deep Learning (DL) for thigh muscle segmentation in MR images holds promise for musculoskeletal architectural assessment, however the process of generating annotated data in supervised approaches is time-consuming.

Goal(s): This study evaluates the impact of scarce annotated data on DL segmentation performance, investigating optimal annotation strategies of thigh muscle MR images.

Approach: Employing thigh MRIs from healthy subjects, the research compares the segmentation performance using various selection strategies and annotated data amount for training a U-Net.

Results: Results reveal high segmentation accuracy (Dice > 0.81) even with minimal annotations (3% of total labels), when selecting the most informative slices for annotation.

Impact: This research highlights the potential of significantly reducing the laborious task of annotating MR images for thigh muscle segmentation, while maintaining robust performance using DL. This efficiency enhancement could expedite the application of DL in muscle health assessment.

Keywords: Head & Neck/ENT, Head & Neck/ENT

Motivation: Fat-suppressed (Fs) contrast-enhanced (CE) three-dimensional (3D) T1-weighted imaging (T1WI) enables the clear visualization of head and neck structures; however, it requires a long scanning time to obtain high quality images.

Goal(s): To demonstrate the utility of model-based deep learning (DL) reconstruction, named SmartSpeed AI, for the acquisition of Fs-CE-3D T1WI of the head and neck.

Approach: Three reconstruction techniques were compared for head and neck Fs-CE-3D T1WI: 1) conventional compressed-sensing sensitivity-encoding (CS), 2) CS followed by end-to-end DL reconstruction, and 3) SmartSpeed AI.

Results: SmartSpeed AI provided the superior image quality than other two reconstruction techniques.

Impact: SmartSpeed AI, a model-based deep learning deep learning reconstruction technique, demonstrated improved image quality in head and neck Fs-CE-3D T1WI, even with a high reduction factor of 12, compared to conventional CS and CS followed by end-to-end deep learning reconstruction.

Keywords: AI/ML Software, Data Processing

Motivation: Magnetic resonance spectroscopy (MRS) is a powerful tool for disease diagnosis, but one of its limitations is the lack of user-friendly processing software or platforms.

Goal(s): Developing an integrated platform that is easy to use, provides powerful hardware, and incorporates advanced processing algorithms.

Approach: CloudBrain-MRS is a cloud-based online platform that has been developed. The platform can be accessed through a web browser without requiring any program installation on the user's side, and it integrates advanced artificial intelligence algorithms.

Results:  The platform supports: 1): Automatic statistical analysis to find biomarkers for diseases;2) Consistency verification between classic and artificial intelligence quantification algorithms; 3)Visualize results.

Impact: This is the first user-friendly cloud computing platform for in vivo MRS with an artificial intelligence processing. The biomedical researchers can do clinical research effectively, and it greatly reduces the requirements for the technical skill of users. 

Keywords: Diagnosis/Prediction, Elastography, Anisotropy, Stiffness, Shear, Tensile, Inversion

Motivation: Most Magnetic Resonance Elastography (MRE) inversion algorithms assume isotropic materials. However, in tissues with a preferred fiber direction, the effective mechanical properties computed under this assumption will reflect a mixture of the true underlying elastic moduli.

Goal(s): In this study, we extend neural network inversion (NNI) to include transversely isotropic (TI) materials. Assumptions are progressively relaxed and the TI inversion in each case is compared against isotropic inversion.

Approach: Data was generated to train and test multiple TI inversions.

Results: An NNI trained to handle TI material can more accurately estimate shear moduli in anisotropic materials and can predict the amount of anisotropy.

Impact: This research expands on currently used MRE to allow for more accurate property estimation in highly organized tissues such as brain and muscle. Moreover, it opens new paths of investigation into pathological changes of the highly organized tissues.

Keywords: Other AI/ML, Machine Learning/Artificial Intelligence

Motivation: MRE is a reliable, quantitative method for the assessment and staging of liver fibrosis. The standard manual MRE image prescription requires proper placement over the liver to ensure consistent MRE quantification. Scan positioning is relatively time-consuming and prone to error and inconsistency.

Goal(s): To develop and implement an automated methodology for MRE prescription from localizers, trained entirely from technologist-prescribed clinical exams.

Approach: Extracted MRE scan coordinates from 354 clinical exams and trained a YOLOv8-nano object detection network to predict prescription planes from a multi-plane localizer series.

Results: We successfully developed a method for automated MRE prescription with implementation on a clinical MRI system.

Impact: Automatic image plan prescription for MRE can minimize technologist-dependent planning errors and scan inconsistency. This may lead to subsequent improvements in both the value and reproducibility of MRE as a quantitative biomarker of liver fibrosis.

Keywords: Other AI/ML, Artifacts, fat suppression, automated detection, cnn

Motivation: Fat suppression on MR images is not always completely successful. Identification of inadequate fat suppression can be difficult for technologists while they are multitasking. Unidentified low quality images inhibit radiologists’ ability to diagnose.

Goal(s): This work develops an automatic method to detect inadequate fat suppression in extremity MRI exams.

Approach: Two-point Dixon fat-water image pairs were combined to simulate varying degrees of fat suppression failure and serve as training data for a CNN.  

Results: Greater than 85% accuracy was obtained on simulated data, which motivates future effort on prospective validation study.

Impact: This work may lead to on-scanner software that auto-identifies inadequate fat suppression, prompting repeat scans with more refined shimming or alternative fat suppression methods. This may improve image quality.

Keywords: AI/ML Software, Machine Learning/Artificial Intelligence, focal liver lesions, Magnetic resonance imaging

Motivation: The number of focal liver lesions (FLLs) detected by imaging has increased worldwide, highlighting the need to develop a robust, objective system for automatically detecting FLLs.

Goal(s): This study aimed to evaluate the application value of deep learning based artificial intelligence (AI) software in detecting FLLs.

Approach: We compared the performance and agreement of deep learning based AI software with those of radiologists in detecting and evaluating malignant lesions in enhanced MRI of patients with FLLs.

Results: AI displayed effective detection performance for malignant lesions down to <10 mm. The measured size of malignant tumors was consistent with the pathologic and manual sizes.

Impact: Our results indicated that the use of AI might promote the detection ability of sub-centimeter-sized liver malignant lesions, providing a reference for selecting clinical treatment schemes.

Keywords: AI/ML Software, Machine Learning/Artificial Intelligence, Contrast enhanced MRA, MRA, Magnetic Resonance Angiography

Motivation: Contrast-enhanced MRA (CE-MRA) of the thoracic aorta is an essential to assess and monitor aortic complications, and to quantify aortic dimensions. However, aortic dimensions’ measurement is cumbersome. Thus, automating aortic 3D-segmentation from CE-MRA is important to improve analysis workflow efficiency.

Goal(s): We aimed to, accurately and precisely, automate thoracic aorta 3D-segmentation from CE-MRA scans using deep-learning.

Approach: Using 125 CE-MRA scans we trained and tested a convolutional neural network to automatically segment the thoracic aortic. 

Results: Automated-segmentations was faster to output and had excellent agreement with manual-segmentations in metrics like aortic diameters and volume, dice scores, Hausdorff distance and average symmetrical surface distance.

Impact: To our knowledge, this is the first study that implemented a fully-automated 3D-segmentation of contrast-enhanced MRA images. Such automation could possibly facilitate the clinical workflow when combined with future applications aiming at automating dimensions’ calculation at standardized locations.

Keywords: AI/ML Image Reconstruction, Cardiovascular

Motivation: Coronary artery calcification score (CACS) is currently a common and widely-accepted indication of UMI, but it itself fails to accurately reflect myocardial ischemia in patients with unrecognized myocardial infarction(UMI).

Goal(s): To establish a UMI-screening workflow for a cohort who receive a physical examination.

Approach: To explore the detection rate of myocardial infarction (MI) using CACS only, Parea only,  CACS in combination with Parea using different thresholds.

Results: The AI-CACS combined with P_area had higher diagnostic performance on differentiating UMI from non-UMI groups than AI-CACS or P_area alone, especially AI-CACS combined with P_area-DL-5SD with AUC of 0.914.

Impact: Patients with UMI usually do not have typical symptoms of cardiogenic chest pain. CACS-P_area-DL-5SD can detect unrecognized myocardial infarction in the outpaitnets, and increased the diagnostic confidence of UMI, providing an important reference for UMI risk stratification and follow-up recommendations.

Keywords: Other AI/ML, Spinal Cord

Motivation: Ability to track real-world performance of AI based spine segmentation models without access to ground-truth data.

Goal(s): Develop AI models which allow prediction of spine vertebrae segmentation model performance in real time.

Approach: Developed a regression and classification deep learning (DL) models that determines quality of segmentation results in terms of Dice overlap metric from a parent segmentation DL model.

Results: For regression model, dice prediction error of 4.3% was obtained, while for categorical classification model, sensitivity between 63-87% observed across evaluation categories. Combination of regression and classification models improves model performance evaluation with sensitivity between 71 to 91%.

Impact: : We developed DL models to automatically evaluate accuracy of spine-vertebrae segmentation models during their deployment in clinical practice without access to ground-truth in both quantitatively (Dice) and qualitatively (Perfect, good, medium, poor). This ensures automatic-logging model effectiveness in real-world data.

Keywords: AI/ML Image Reconstruction, Image Reconstruction, cross-frequency B1 prediction

Motivation: Coil sensitivity profiles are essential for multi-channel MRI/MRSI data processing, yet not acquirable within reasonable scan time for X-nuclei with low natural abundance.

Goal(s): Predict sensitivity patterns of lowly abundant X-nuclear species based on sensitivities of highly abundant nuclei acquired by the same multi-tuned coil array.

Approach: We scanned 8 subjects at 1.5T and 3T using similar commercial head arrays. A 3D patch-based convolutional neural network is used to predict 3T sensitivity patterns from 1.5T sensitivities. 

Results: Predicted 3T sensitivity patterns show high similarity to the ground truth. 3T signal-combination is feasible using the 1.5T-based predicted sensitivities, despite subject repositioning and hardware deviation.

Impact: An adequate prediction of coil sensitivity profiles at 128MHz based on 64MHz sensitivity profiles using highly similar receiver arrays was achieved. It opens up new possibilities for combining multi-channel signals acquired by multi-tuned (e.g., ³¹P-²³Na, ¹⁹F-¹H, etc.) receiver arrays.