Keywords: Cancer, Radiomics

Motivation: In rectal cancer, the existing stratification of prognosis is mainly based on the established TNM tumor staging system, which limits clinical decision-making.

Goal(s): Construct a model to provides prognostic information preoperatively.

Approach: This study established radiomics models  to predict the disease-free survival (DFS) in rectal cancer, based on tumor, and multiple body components (including mesenteric fat and pelvic skeletal muscles) in MRI.

Results: The radiomics model based on tumor and multiple body components of MRI in rectal cancer, have good predictive value for DFS in rectal cancer patients at 2 years after surgery.

Impact: The model that provides prognostic information at the time of cancer diagnosis would be useful for optimizing treatment and monitoring.

Keywords: Kidney, Kidney, ASL, 3D Cartesian TSE, CS-AI, Deep Learning

Motivation: Motion and breathing can compromise 3D renal ASL MRI, reducing SNR and causing artifacts. Shorter-time acquisition is crucial for its clinical utility

Goal(s): To Evaluate the potential of CS-AI in accelerating renal ASL MRI with 3D Cartesian TSE. 

Approach: We accelerated renal ASL MRI 2-, 4-, and 6 times using CS-AI, comparing with SENSE. 

Results: CS-AI-accelerated images exhibited superior SNR and quality compared to SENSE, without affecting RBF quantification.

Impact: This study may enhance the clinical utility of 3D renal ASL, particularly in discerning perfusion alterations in small-sized lesions like small renal masses.

Keywords: Kidney, Genetic Diseases, ADPKD

Motivation: Height-adjusted total kidney volume (htTKV) growth rate measured on MRI or CT is a critical biomarker for monitoring autosomal dominant polycystic kidney disease (ADPKD) progression.

Goal(s): This study aims to develop a tool for accurate calculation of the htTKV growth rate based on all available MRI scans.

Approach: Accuracy of four MRI methods for calculating htTKV growth rate were assessed as compared to ground truth calculated from 10+ years of imaging follow up.

Results: Using 2-parameter least squares fitting with 3+ scans or 5+ years of follow up reduce error 2-fold compared to the current clinical standard, Mayo Imaging Classification.

Impact: Accurate estimation of kidney growth rate on abdominal MRI using FIT will enable better prediction of disease progression and response to therapy in patients with ADPKD.

Keywords: Lung, Lung

Motivation: Spiral-UTE MRI has been proposed for more efficient lung imaging to permit breath-hold ultra-short echo time acquisition of the lung. It is more valuable to further accelerate the acquisition of the spiral-UTE MRI of lung images, thus enabling shorter breath-holds and higher spatial resolutions.

Goal(s): This work presents a deep learning based method to enable the reconstruction of spiral-UTE MRI of lung images from accelerated spiral k-space.

Approach: An unrolled network was developed for reconstructing images from the accelerated non-cartesian k-space.

Results: The unrolled network allows for higher reconstruction quality for spiral-UTE MRI of lungs compared to a standard U-Net.

Impact: The proposed unrolled network tailored for spiral MRI reconstruction enables reconstruction of accelerated spiral-UTE of lung images to allow shorter breath-holds and higher spatial resolutions. This reconstruction technique can also extended to other multi-coil non-cartesian accelerated MRI reconstructions.

Keywords: Lung, Lung

Motivation: Ultrashort echo time (UTE) MR imaging has been developed to visualize lung tissues, but is limited by  long acquisition time and capacity for displaying pulmonary tissues. 

Goal(s):  3D free breathing stack-of-spirals UTE (spiral-UTE) could provide a higher readout efficiency, but its comprehensive performance  with and without contrast material for follow-up of patients with identified malignant tumors at 1.5T has not been reported. 

Approach: Aim to investigate  clinical feasibility of spiral-UTE  compared to enhanced-VIBE with CT as the reference.

Results: Spiral-UTE was superior with improved image qualities for depicting bronchi and lung parenchyma  than enhanced-VIBE in lung screening during the oncology patient follow-ups

Impact: Spiral-UTE provides a potential alternative to CT for lung follow-up with a significantly superior quality visualization of the pulmonary anatomy than routine enhanced-VIBE

Keywords: Lung, Lung, Zero echo time, pulmonary ventilation

Motivation: Lung MRI using UTE and ZTE techniques is limited by its SNR and tissue interface blurring. Deep learning based reconstruction (DLR) technique has been used to improve MRI image quality via noise reduction.

Goal(s): To evaluate potentially clinical applications of breath-hold DLR ZTE lung MRI in ventilation function.

Approach: DLR and conventional reconstructed ZTE lung images of thirty patients with pulmonary nodules were compared for image quality and image-based pulmonary ventilation estimation. 

Results: Compared to conventional reconstructed results, DLR ZTE images demonstrated improved image quality and better correlation with clinical measurements for ventilation estimation. 

Impact: This preliminary study demonstrated the feasibility of DLR ZTE technique in lung MRI. DLR ZTE images showed improved image quality and better correlation with clinical measurements for ventilation estimation. 

Keywords: Lung, Lung

Motivation: There is a lack of non-invasive approaches for quantitatively analyzing the patterns of respiration motion in proton MRI in patients with lung diseases such as post-COVID symptoms.

Goal(s): The goal of this work is to determine whether post-COVID-19 patients can be classified as having either long COVID or no symptoms by analyzing dynamic MRI motion fields within various regions of lungs.

Approach: A deep learning-assisted framework was developed for automatically analyzing localized respiratory motion in lung MRI.

Results: The framework was able to successfully categorize patients into different categories based on their degrees of no symptoms using the proposed image analysis framework.

Impact: This work develops an automated framework that can aid radiologists in quickly determining not only the presence but also the severity of long COVID. It can also be extended for applications in other lung diseases.

Keywords: Prostate, Machine Learning/Artificial Intelligence

Motivation: Radiologists face challenges in the accurate prediction of prostate cancer (PCa) with gray-zone PSA levels. Deep learning (DL) evaluated PCa with gray-zone PSA levels remains unclear.

Goal(s): The aim of this work was to investigate the comparative performance of DL and radiologists. We trained a 3D DenseNet 121 model for automatic PCa classification with gray-zone PSA levels.

Approach: We trained a 3D DenseNet 121 model for automatic PCa classification with gray-zone PSA levels.

Results: The DL model yielded an AUC of 0.95 (0.85-1.0) for the identification of PCa with gray-zone PSA levels in the test set, significantly improving performance over the inexperienced radiologists.

Impact: The deep learning model yielded an AUC of 0.95 (0.85-1.0) for the identification of PCa with gray-zone PSA levels in the test set, significantly improving performance over the inexperienced radiologists. 

Keywords: Prostate, Prostate, k-space data, public

Motivation: There is a lack of publicly available, raw k-space data for prostate MRI.

Goal(s): To compile and release raw k-space data for clinical prostate MRI and demonstrate its utility for development of deep learning methods for image reconstruction and automated diagnosis.

Approach: Biparametric MRI data from 312 patients with associated prostate cancer labels were added to the public fastMRI repository. Deep-learning models were trained on the data to reconstruct images from undersampled k-space and perform automated diagnosis of prostate cancer (PCa) on these images.

Results: SSIM > 0.866 and AUC > 0.80 (test set) for the deep-learning reconstruction and automated PCa diagnosis respectively.

Impact: Raw k-space data with clinical labels from fastMRI prostate will enable researchers to develop clinically relevant deep-learning reconstruction and automated diagnosis models which may ultimately advance the diagnosis and management of prostate cancer.

Keywords: Prostate, Metabolism, Metabolomics, NMR

Motivation: Current diagnostic methods cannot predict the aggressiveness of prostate cancer (PCa) at a treatable stage of  disease. 

Goal(s): To interrogate tumorigenesis of PCa and AI/ML techniques to NMR-based targeted blood plasma metabolomic profiling analysis for prediction of PCa. 

Approach: Use AI/ML approaches to NMR metabolic profiling for PCa patient blood plasma data analysis 

Results: Phosphocreatine, choline, 3-hydroxybutyrate, taurine and glucose showed highest discriminate using CFS, PLS-DA, OPLS-DA,  random forest models. 

Impact: It will pave way to enhance understanding of cancer pathogenesis and biomarker/s identification and early detection systems.

Keywords: Liver, Liver

Motivation: Liver T2-weighted imaging usually requires a long scan time. A faster sequence with adequate image qualities is essential in clinical practice.

Goal(s): To evaluate deep-learning reconstruction accelerated T2-weighted HASTE sequence in liver application on the image quality and diagnostic confidence.

Approach: One hundred and five patients were imaged using both sequences. Images were reviewed independently by two blinded observers.

Results: The DL HASTE sequence can detect more liver lesions and improve the CNR of the lesion compared to the conventional T2-weighted BLADE sequence, with a 2.5-fold reduction in acquisition time.

Impact: DL HASTE sequence has the potential to replace the conventional BLADE sequence in routine clinical liver MRI, reducing the scan time and detecting more liver lesions.

Keywords: Liver, Radiomics, Liver fibrosis; Gd-EOB-DTPA-enhanced MRI; nomogram

Motivation: Despite the effectiveness of elastography and serology tests in detecting liver fibrosis, diagnosing early-stage fibrosis remains challenging.

Goal(s): Develop and validate a reliable radiomics model using Gd-EOB-DTPA-enhanced MRI for early liver fibrosis diagnosis.

Approach: Create a radiomics model based on Gd-EOB-DTPA-enhanced MRI and establish a fused nomogram combining clinical characteristics and LSM. Compare the diagnostic performance of the fused model with single models for early-stage liver fibrosis.

Results: Gd-EOB-DTPA-enhanced MRI radiomics model effectively diagnoses early liver fibrosis. The fusion model enhances diagnostic efficiency.

Impact: To develop and validate a fusion model based on Gd-EOB-DTPA-enhanced MRI to identify early-stage liver fibrosis.

Keywords: Liver, Liver, LAVA-Flex, 3D FLEX DL

Motivation: Fat suppressed T1 images, such as LAVA-FLEX, are routinely used in liver imaging, but can suffer from SNR and IQ issues. 

Goal(s): Our goal was to validate application of 3D deep learning to 3D LAVA-FLEX in routine adult liver imaging via a reader study and noise characterization. 

Approach: DL and conventionally reconstructed images were assessed across several IQ attributes (motion, ringing, edge, vessel) by two radiologists. Noise characteristics were evaluated by calculation of total variation and edge detection. 

Results: Based on the calculated odds ratios, the radiologists preferred DL across the various IQ attributes, with decreased noise and improved sharpness in DL images. 

Impact: The application of 3D DL to routine 3D LAVA-FLEX imaging provides increased diagnostic quality, and has the potential to improve routine abdominal care in patients who can't hold their breath.

Keywords: Liver, Liver, TACE refractoriness，HCC，MRI，Radiomics

Motivation: Parts of HCC  respond poorly to TACE, and its efficacy declines as the number of procedures increases in the clinical practice, which is called TACE refractoriness.

Goal(s): To develop and validate a preoperative multiparametric MRI-based radiomics model to predict TACE refractoriness in HCC.

Approach: Radiomics feature selection was performed using PCC and RFE, and SVM was used to construct radiomic models based on each sequence and their combination, clinical-radiological model based on selected clinical-radiological predictor and combined model.

Results: The combined model exhibited excellent predictive performance. The multi-phase radiomics signature performed better in predicting TACE refractoriness compared to the best single-phase radiomics signature.

Impact:  The preoperative multiparametric MRI radiomics analysis can predict TACE refractoriness in hepatocellular carcinoma, which may provide better guidance for decision-making regarding further TACE treatment and optimize the mode of treatment and patient management, ultimately resulting in patient survival benefits.

Keywords: Liver, Cancer, Habitat imaging

Motivation: Isocitrate dehydrogenase 1 (IDH1) mutation is an important therapeutic target for intrahepatic cholangiocarcinoma (ICC).

Goal(s): To achieve non-invasive prediction of IDH1 mutation in ICC to assist in clinical management.

Approach: Preoperative Gd-EOB-DTPA-enhanced MRI features and clinical information were retrospectively collected. Habitat analysis was performed based on pre- and post-enhancement T1 maps. Nomogram prediction model was established based on filtered variables.

Results: Higher serum AFP level, higher T1 change ratio, more frequent intratumoral vessel and T2 central brightness, and habitat 5 are risk factors for IDH1-mutated ICC. The combined nomogram model demonstrated the highest diagnostic performance over the clinilal+imaging model and the habitat model.

Impact: The proposed strategy, Gd-EOB-DTPA-enhanced MRI and T1–based habitat imaging, can be applied for preoperatively and noninvasively identifying IDH1-mutation status in ICC, which offers potential benefits in terms to aid in clinical management.

Keywords: Liver, Multi-Contrast, Image-to-image translation

Motivation: Either fast 2D T2-weighted abdominal imaging or 3D T2 MIP techniques have limitations. There remains a need for fast 3D T2 abdominal high-resolution imaging.  

Goal(s): To develop a conditional GAN model to synthesize T2-weighted images from 3D high-resolution T1-weighted abdominal images preserving spatial resolution of the source images.

Approach: Abdominal images acquired from 39 volunteers were included for the study. A conditional GAN model was trained to generate T2-weighted images from T1-weighted images slice by slice.

Results: Overall, the generated T2-weighted images were similar to the real T2-weighted images, though some contrast differences in the bowels and kidneys were seen. 

Impact: This proof of principle study shows the GAN model can be used to generate T2-weighted images from T1-weighted images, with the potential for rendering high quality volumetric 3D high-resolution abdominal T2-weighted images that is superior to current 3D MIP methods.