Keywords: Liver, Segmentation, ADPKD, PLD, Liver Cyst, Deep Learning, Segmentation Model

Motivation: Autosomal dominant polycystic kidney disease (ADPKD) often also has polycystic liver disease (PLD), impacting patients' well-being. Manually segmenting liver cysts for measuring disease burden is time-consuming and error-prone, necessitating an automated solution.

Goal(s): We introduce a deep-learning (DL) framework for liver cyst segmentation in ADPKD/PLD patients.

Approach: An nnUNet-based framework ensembled 2D and 3D models trained on our institute's ADPKD dataset to detect liver cysts in an external test set. Additionally, we implemented patient, cyst, and voxel-level evaluation metrics for clinical impact assessment.

Results: Our model achieved an 84% cyst-level Dice score significantly reducing annotation time by 91%.

Impact: This research aims to revolutionize PLD monitoring by transitioning from qualitative to quantitative, replicable, and scalable approaches. Advanced DL models can produce high-quality liver cysts annotations and introduce cyst-level evaluation metrics, aiding radiologists with precise disease assessment and clinical decisions.

Keywords: Liver, Diffusion/other diffusion imaging techniques

Motivation: Single-shot and multi-shot M1-optimized diffusion imaging (MODI) are recent DWI methods used to mitigate motion and distortion artifacts, yet they often experience chemical shift-based fat suppression failures in the abdomen.

Goal(s): To optimize fat suppression in multi-shot MODI-DWI of the abdomen.

Approach: Slice-specific chemical shift-encoded (CSE) data-informed optimization of shims is combined with single-shot and multi-shot MODI-DWI in 7 subjects imaged at 3T.

Results: Improved fat suppression and water signal excitation were observed alongside the motion and distortion artifact reduction provided by multi-shot MODI-DWI. Unwanted fat signal was reduced through this technique in areas of interest such as the liver, spleen, and ribcage.

Impact: Motion-robust, low-distortion DWI of the abdomen, with reliable fat suppression is demonstrated by combining multi-shot EPI, M1-optimized DW waveforms, and an optimized slice-by-slice shimming approach. This combined method may enable improved detection and staging of cancer in the abdomen.

Keywords: Cancer, Machine Learning/Artificial Intelligence, Image Quality; Quality Control / Quality Assurance; QA/QC; AI/ML image reconstruction

Motivation: With increasing AI adoption in MR-reconstructions, robust quality assessment becomes paramount. This study aims to ensure that AI-techniques meet clinical requirements at implementation and longitudinally.

Goal(s): 1) Compare image quality of AI-imaging with standard techniques in anorectal cancer. 2) Develop longitudinal quality control (QC)  assessments capable of detecting changes in AI-reconstructions without resource-intensive evaluations.

Approach: A prospective study involving 40 patients utilised radiologist scoring and quantitative image-quality-metrics (IQMs). Retrospective reconstructions gauged sensitivity of IQMs to reconstruction pipeline changes.

Results: AI-reconstructions demonstrated >50% time savings with improved image quality. Feasibility of quantitative-IQMs for assessing AI-reconstructions is established, providing a practical solution for ongoing QC.

Impact: There is a need to develop QC assessments offering performance monitoring for AI-based reconstructions in diverse clinical settings. The study presents feasible ways to support integration of AI-imaging into clinical practice, including resource-efficient quantitative image quality assessments. 

Keywords: Liver, Liver

Motivation: Gd-EOB-DTPA-enhanced hepatobiliary phase (HBP) imaging is clinical routine for liver lesion identification, and is usually empirically conducted at 20 minutes after bolus injection.

Goal(s): Our goal was to demonstrate the feasibility of optimizing clinical workflow by synthesizing Gd-EOB-DTPA-enhanced HBP images via machine learning.

Approach: Precontrast and early-enhanced T1WIs (5-min after bolus injection) acquired at 3 T were used to synthesize HBP images via a generative adversarial network in 490 subjects.

Results: Our preliminary results showed that synthesized HBP images are visually comparable to acquired HBP images with high SSIM(0.87±0.08) and PSNR(29.6±2.25).

Impact: Machine learning synthesized HBP images could provide comparable diagnostic information to acquired HBP images, suggesting that machine learning might be used to optimize clinical workflow and greatly shorten acquisition time for Gd-EOB-DTPA-enhanced MRI. 

Keywords: Pelvis, Multimodal

Motivation: The consensus molecular subtype (CMS) is a novel classification system that reflects the genetic characteristics of the tumor. CMS4 is associated with the worst prognosis.

Goal(s): To investigate whether a radiomics-based machine learning approach could predict CMS4 status in CRC patients.

Approach: The sequencing data was input into the CMS classification system to generate CMS subtype outcomes. Radiomics features were extracted from baseline T2WI and contrast-enhanced MRI. Machine learning algorithms were applied  to explore the best-performing and most robust model. 

Results: The best performing model achieved AUCs of 0.855 and 0.759 in the test set and external validation set.

Impact: The genetic phenotype of CMS4 colorectal cancer may be potentially associated with morphological features. Multiparametric radiomics-based machine learning shows promising potential in distinguishing CMS4 from other subtypes of CRC.

Keywords: Pelvis, Machine Learning/Artificial Intelligence

Motivation: The promising application of deep learning (DL) techniques for prognostic prediction in various tumors has been reported, but mostly with single-task models

Goal(s): Exploring the use of multi-task DL models to automate the whole process of prediction for rectal cancer patients.

Approach: We designed a modality-fusion-based multi-task DL model to concurrently predict tumor volumes, patient relapse state, and patient risk scores based on a combination of multimodal MR images and clinical tabular data.

Results: The multi-task DL model achieved favorable predictive performance at the stage of initial diagnosis with automatic lesion identification, and further improved with the inclusion of postoperative pathology indicators.

Impact: Multi-tasking DL may be a new approach and orientation to fully automate the process of clinical prediction, and its feasibility is expected to be further explored in other oncology studies in the future.

Keywords: Breast, Breast, Synthetic MR,Deep Learning based reconstruction

Motivation:  Synthetic MRI, with its unique advantages including unique signal acquisition, rapid synchronization, visualization and multiparameter maps, is gradually applied in breast cancer diagnosis. However, its extended scanning time restricts its broader use.

Goal(s): To accelerate synthetic MRI while maintaining its quantitative parameters and image quality using deep learning-based reconstruction (DLR).

Approach: 12 female patients were enrolled and scanned with two sets of synthetic MRI: a standard protocol and an accelerated protocol (before and after DLR). Quantitative parameters, SNR of lesion and subjective image quality were compared.

Results: Comparable image quality was achieved using accelerated synthetic MRI with DLR.

Impact: The combination of DLR with accelerated synthetic MRI protocol has significant benefits in promoting the practical application of synthetic MRI in breast imaging and enhancing examination efficiency.

Keywords: Liver, Cancer

Motivation: Identifying MVI before surgery is crucial to optimizing treatment strategies and predicting prognosis.

Goal(s): To develop a multi-region radiomics-based predictive model using multi-sequence MRI to assess MVI in HCC.

Approach: Three Models (Tumor, Tumor-Expand5, Tumor-Expand10) were constructed and evaluated.

Results: The Tumor-Expand5 model, with the best predictive accuracy, demonstrates its effectiveness. The research explores the advantages of 5mm VOI expansion for MVI diagnosis over the 10mm VOI expansion, providing valuable insights. Despite limitations, this study offers a preoperative tool to predict MVI in HCC, enhancing clinical decision-making.

Impact: Predicting Microvascular Invasion (MVI) in HCC through multi-region radiomics-based MRI advances precision medicine and treatment optimization. The Tumor-Expand5 model's superior diagnostic performance demonstrates the potential to enhance clinical decision-making.

Keywords: Cancer, Cancer, Rectal cancer, Deep learning radiomics, Magnetic resonance imaging, Synchronous liver metastasis

Motivation: Accurate synchronous liver metastasis (SLM) risk stratification is important for treatment planning and prognosis improvement. 

Goal(s):  Our goal is to establish a non-invasive and quantitative prediction model of synchronous liver metastases (SLM) in rectal cancer (RC) to help with accurate staging.

Approach:  The deep learning (DL) model was fitted based on multi-parameter MRI of primary cancer combined with Clinical features (CF) features, and 5-fold cross-validation and external validation were performed.

Results: We demonstrated that the combination of CF and DL features achieved a satisfactory predictive performance for SLM, and also confirmed the generalizability of this model by external validation.

Impact: The discovery of the DL model  would change treatment strategies. For patients with high-risk metastasis, a more aggressive systemic examination and shorter follow-up should be considered and may contribute to improved outcomes.

Keywords: Lung, Hyperpolarized MR (Gas)

Motivation: Xenon ventilation MRI shows distinct defect patterns that appear disease-specific but are difficult to measure. Deep learning, via neural networks, can generate texture features to classify images. Image classification has applications in diagnostics, phenotyping and predicting outcomes.

Goal(s): To determine if neural networks could determine disease classification from xenon MRI.

Approach: 2D neural networks were trained on data from eight disease states (including healthy controls) and assessed on top-1, top-3 accuracy and recall.

Results: The top performing network had a 54% top-1 and 86% top-3 accuracy.

Impact: Artificial intelligence can classify disease from xenon MRI alone with moderate accuracy and differentiate between similar conditions. In the future, deep learning could be used diagnostically, for phenotyping disease subgroups and predicting outcomes.

Keywords: Breast, Machine Learning/Artificial Intelligence, Breast Imaging, Multiplexed Sensitivity Encoding (MUSE), Diffusion Weighted Imaging

Motivation: Diffusion-weighted imaging (DWI) in breast imaging is constrained by image distortion, which can be mitigated through the utilization of multi-shot DWI (MUSE). 

Goal(s): We conducted a pilot study to investigate the impact of deep-learning reconstruction (DLRecon) on MUSE image quality. 

Approach: Compared with the non-DL MUSE images, the MUSE DLRecon showed higher SNR without affecting the mean ADC value. Moreover, employing a higher shots in MUSE DL with reduced NEX could provide less-distortion DWI. 

Results: Our preliminary results suggest the feasibility of MUSE-DWI in breast imaging with a higher number of shots.

Impact: Our results suggest that the DLRecon could be beneficial for the regions prone to distortion and requiring a high density of diffusion direction information, in the complex diffusion modeling, all while maintaining a feasible scan time in breast MUSE imaging.

Keywords: Breast, Breast

Motivation: EPI-based DWI suffers from ghosting, chemical shift, and distortion artifacts. FSE-based DW-PROPELLER has been shown to overcome the above artifacts but at the cost of longer scanner time.

Goal(s): To evaluate the combination of DW-PROPELLER with a deep learning (DL)-based reconstruction to provide motion-robust distortion-free high spatial resolution breast DWI.

Approach: Phantom and in-vivo breast images were acquired using DW-PROPELLER followed by both conventional and DL reconstruction.

Results: DW-PROPELLER with DL showed less distortion, less chemical shift artifacts, and increased SNR and sharpness compared with multi-shot DW EPI in both phantom and in-vivo breast imaging.

Impact: This work demonstrated the feasibility of using a deep learning-based approach to improve image sharpness, reduce noise, and chemical shift artifacts for motion-robust and distortion-free high spatial resolution diffusion-weighted breast imaging.

Keywords: Breast, Cancer, deep learning; super resolution; screening

Motivation: High-resolution images in breast MRI are desired for lesion detection and characterization but are restricted due to scan time constraint in routine clinical settings.

Goal(s): Our goal was to use deep learning (DL)-based reconstructions to improve image resolution and quality of routine clinical breast MRI.

Approach: We applied a dedicated Precise-Image-Net for both 2D T1- and T2-weighted imaging in breast cancer patients at 1.5T and compared it to conventional parallel imaging, compress sensing, and convolutional neural network (CNN) reconstructions.

Results: Initial clinical data demonstrated a clear improvement of sharpness in breast T1- and T2-weighted images compared with standard reconstructions.

Impact: Deep learning-based super-resolution reconstruction provides improved image resolution and sharpness in breast MRI, showing promises for better lesion detection and characterization in routine clinical settings without prolonging scan time, which is of particular importance in dynamic contrast enhanced-MRI.

Keywords: Prostate, Body

Motivation: The motivation of this study is to develop novel methods for mapping non-invasively the underlying history-morphometric features of prostate cancer.

Goal(s): The goal of this study is to develop and demonstrate radio-pathomic mapping techniques to enable noninvasive detection of prostate cancer presence and distinction from benign tissue using MRI.

Approach: Our approach was to align multiparametric MRI with digitized histology slides from prostatectomy specimens, then predict quantitative histological features from MRI intensities, and use these predicted features to classify cancer versus noncancer regions.

Results: Our models can distinguish cancerous from noncancerous prostate tissue with 70% accuracy.

Impact: This study uses radio-pathomic mapping for noninvasive prostate cancer detection, demonstrating the potential to differentiate cancerous vs benign prostate tissue using imaging surrogates of microstructural features discernible only on histology.

Keywords: Prostate, Prostate

Motivation: Low-Field MR offers a great platform for low-cost high-performance screening of prostate cancer, but it suffers from low SNR. Prolonged scan times are typically needed to achieve adequate SNR at low field. 
 

Goal(s): In this work, we developed an advanced deep learning denoising method for rapid high spatial resolution prostate MRI at 0.55T. 

Approach: The proposed approach was tested in T2-weighted prostate MRI. Supervised training was performed to denoise images acquired with different numbers of averages, corresponding to different scan times. 

Results: Deep learning was able to denoise prostate images at high spatial resolution  resulting acquisition time with 1-2 average.

Impact: The proposed denoising technique holds significant potential to promote the use of 0.55T MRI and other types of low-field MRI for prostate imaging and screening for prostate cancer, with reduced cost and greater accessibility. 

Keywords: Kidney, Kidney, Segmentation, TKV, Transformers

Motivation: Convolutional Neural Networks (CNNs) have long been the go-to deep-learning architecture for medical image segmentation, but in recent years transformer-based architectures adapted from large language models are setting a new standard. 

Goal(s): The aim of this study was to test if transformers are suitable for 3D kidney segmentation on high-resolution MRI. 

Approach: A transformer-based deep-learning architecture (UNETR) was trained and tested against a supervised method on 82 patient datasets from the iBEAt study on diabetic kidney disease. 

Results: UNETR provides fast segmentation with comparable results to the supervised method, but additional refinement is needed to reduce the limits of agreement.

Impact: Novel transformer-based architectures for medical image segmentation may be useful for fast 3D segmentation of individual kidneys.

Keywords: Fetal, Brain, Data Analysis, Data Process, Image Reconstruction

Motivation: Fetal MRI is important in clinical and scientific applications but prone to motion artifacts. Automated image quality assessment (IQA) assists data acquisition and subsequent analyses. However, training neural networks for IQA requires labor-intensive manual annotation.

Goal(s): To develop a model for fetal MRI IQA that doesn't require image quality labels.

Approach: A network is trained to determine the acquisition orientation of 2D T₂-weighted images.  The variation of orientation recognition network (ORN) inferences for central images of a brain stack is used to assess motion and the image quality.

Results: High-quality and low-quality images are robustly discriminated. Image super-resolution from brain stacks is improved.

Impact: ORN-IQA eradicates the necessity image quality labels for training, thereby circumventing manual annotation. ORN-IQA simplifies online image quality evaluation and permits image reacquisition during fetal MR scans. Moreover, ORN-IQA improves super-resolution reconstruction results.

Keywords: Liver, Image Reconstruction

Motivation: Free-breathing T2 mapping of the abdomen is possible for subjects that cannot hold their breath, but current techniques require long scan times that are not always possible.

Goal(s): To produce high quality T2 weighted images and T2 parameter map from highly accelerated scans of the abdomen.

Approach: Combining the radial turbo spin echo sequence, navigator triggering, and new deep learning reconstruction techniques to increase the acceleration of the acquisition while maintaining image quality.

Results: Using these techniques, one can produce high quality T2 weighted images and T2 parameter map of the entire abdomen in under 4 minutes.

Impact: Deep learning techniques significantly reduce both scan time and reconstruction time for highly accelerated, navigator-triggered free breathing T2 weighted images and T2 parameter map of the abdomen.

Keywords: Breast, Cancer, Deep Learning Reconstruction, Diffusion-Weighted Imaging, Breast Diagnosis, Synthetic MRI

Motivation: Breast cancer has emerged as the foremost global malignancy, prompting a growing inclination toward exploring novel non-invasive imaging techniques that obviate the need for contrast agent administration.

Goal(s): Enhancing breast diagnostics without reliance on contrast agents.

Approach: Expanding on the foundation of deep learning-based DWI reconstruction, coupled with Synthetic MRI, as a viable alternative to traditional contrast-enhanced diagnostic methodologies, the focus lies in pinpointing valuable parameters for differential diagnosis.

Results: The fusion of deep learning-reconstructed DWI and Synthetic MRI yields an impressive AUC (Area Under the Curve) of 0.995 in distinguishing between benign and malignant breast pathologies.

Impact: The integration of deep learning-reconstructed DWI with Synthetic MRI not only carries substantial diagnostic significance in discerning between benign and malignant breast conditions but also exhibits the promise of supplanting conventional contrast-enhanced methodologies.