Keywords: Diagnosis/Prediction, Bone, Osteosarcoma

Motivation: To compare the efficacy of radiomics models using four machine learning algorithms in predicting for local relapse of osteosarcoma before surgery.

Goal(s): This study established a robust prediction model of local relapse to improve prognosis efficacy of osteosarcoma and aid a personalized treatment planning.

Approach: Comparison of four algorithms in classifying high-risk local-relapse patients from low-risk ones based on only preoperative radiographic, MR, and both radiomic features.

Results: The random-forest based prediction model using both radiograph-MRI radiomic features had the best performance on differentiating patients with local relapse from those with non-local relapse with AUC of 0.868, sensitivity of 0.909, specificity of 0.750.

Impact: This study facilitated early identification of high-risk local-relapse osteosarcoma patients who may benefit from model-guided therapeutic interventions and have better long-term outcomes.

Keywords: Diagnosis/Prediction, Infection, spondylitis

Motivation: Tuberculous spondylitis （TS） and brucellosis spondylitis （BS） are two common infectious diseases in spinal surgery, and the differential diagnosis of these diseases is challenging but important to ensure appropriate treatment.

Goal(s): The aim of this study was to evaluate the performance of a convolutional neural network CNN) based on VGG19 in distinguishing between TS and BS on different parameter magnetic resonance imaging （MRI） and to compare it with three radiologists.

Approach: MRIs of 383 patients were randomly divided into training （n = 307） and validation （n = 76） groups.

Results: VGG19-based CNN outperforms radiologist assessment in distinguishing TS from BS.

Impact: The proposed CNN based on VGG19 is effective in diagnosing TS and BS on MRI, which could not only help in clinical decision making, but also improve efficiency and reduce medical costs.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Machine learning; Cardiac magnetic resonance; Multi-modality; Dilated cardiomyopathy; Left ventricular reverse remodeling

Motivation: Few multi-modality machine learning (ML) classifiers combine cardiac magnetic resonance (CMR) imaging with clinical data for predicting LVRR in DCM patients, limiting improvements in patient outcomes and management.

Goal(s): To develop an ML classifier using multi-modality data, including CMR, to predict LVRR in initial DCM patients.

Approach: 129 DCM patients with complete clinical and CMR data were collected. Feature selection identified relevant variables, and an LR-based nomogram was constructed and evaluated. 

Results: The nomogram achieved an AUC of 0.857 in the test cohort, incorporating late gadolinium enhancement pattern, global longitudinal peak strain, aldosterone antagonist, and severe mitral regurgitation.

Impact: The CMR-based multi-modality nomogram has a superior ability in the prediction of LVRR in DCM patients.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: Deep learning for myocardial scar segmentation offers an alternative to time-consuming and observer-dependent semi-automatic approaches.

Goal(s): The objective of this study was to assess the impact of effective image resolution on neural network training for ventricular scar segmentation.

Approach: Convolutional neural networks were trained on magnetic resonance images with constant matrix size and field-of-view but differing resolutions, and tested on a range of resolutions to investigate the effects.

Results: Neural networks trained on a specific resolution indicated a bias of the scar area estimation when employed to lower -or higher-resolution images. Deploying a network trained on multiple resolutions resulted in reduced resolution dependency.

Impact: The effective image resolution, with constant matrix size and field-of-view, should be considered when training a segmentation model to alleviate unwanted bias in the estimation. Training on multiple resolutions has been shown to increase network precision and robustness.

Keywords: Diagnosis/Prediction, Segmentation

Motivation: Undiagnosed aortic aneurysms can be fatal. We aim to use machine learning to measure the aorta from standard CMR localisers, allowing screening and characterisation of aneurysms without the need for additional sequences.

Goal(s): We aim to generate accurate 3D segmentations (1-1.5mm slice thickness) from standard 2D trans-axial SSFP localisers stacks (10-12mm slice thickness).

Approach: We trained an AI model using high-resolution segmentations alongside simulated low-resolution images (2D localisers). This enables the model to predict high-resolution segmentations from unseen, low-resolution images by generalising from the learned patterns.

Results: Our model shows promising performance in generating high-resolution segmentations from various unseen low-resolution validation dataset.

Impact: With our model, the dilated aorta can be identified from routine CMR scans without the need for extra sequences. Additionally, 3D aorta morphology information can be obtained from previous clinical CMR studies or population studies without additional cost.

Keywords: Diagnosis/Prediction, Inflammation, Relaxometry

Motivation: Conventional methods for imaging inflammation-related bone marrow (BM) changes are limited, necessitating advanced quantitative MRI approaches. However, capturing microscopic details poses challenges. Existing T2 distribution estimation methods' limitations led to the development of P2T2-Boot.

Goal(s): We aimed to improve BM inflammation analysis using P2T2-Boot and assess its ability to differentiate healthy and inflamed BM.

Approach: P2T2-Boot, a neural network for T2 distribution estimation, was developed with bootstrapping techniques, trained on simulated MRI signals, and tested on real mice data.

Results: The bootstrapped model outperformed others at low Signal-to-Noise Ratios and demonstrated superior performance in distinguishing inflammatory and non-inflammatory mice.

Impact: P2T2-Boot significantly enhances detecting BM inflammation, excelling in noisy conditions. Its superiority underscores its potential for advancing disease studies. The method's potential extensions make it a promising tool for advancing inflammation-related disease studies and clinical applications.

Keywords: Diagnosis/Prediction, Hypertension

Motivation: Right heart catheterization (RHC) is the invasive gold standard to measure several cardiopulmonary parameters.  PREFUL is a non-invasive method to quantify hemodynamics during the cardiac cycle from MRI in free breathing and without contrast agents.  

Goal(s): Provide an MRI-based method for non-invasive estimation of mean pulmonary arterial pressure (mPAP), pulmonary arterial wedge pressure (PAWP) and mixed venous oxygen saturation (SvO₂) currently determined by RHC.  

Approach: Multiple deep neural networks were trained to estimate mPAP, PAWP and SvO₂ from cardiac cycles of three lung slices, provided by PREFUL. 

Results: The estimations of mPAP, PAWP and  SvO₂ showed strong correlation with RHC. 

Impact: Estimation of mPAP, PAWP and SvO₂ is possible via  PREFUL MRI. The presented approach is more automatized than echocardiography and may support the diagnosis of cardiopulmonary diseases such as pulmonary hypertension where RHC is not yet available.  

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: The need to address the challenge of a high nonresponse rate (approximately 40%) in CRT patients.

Goal(s): By leveraging advanced computational methods, this research seeks to redefine risk stratification and long-term survival prediction.

Approach:

• 3D-CAE model is designed to compress displacement trajectories into a low-dimensional latent code while preserving sufficient information for trajectory reconstruction.
• The survival network utilizes latent features from three specific slices for predicting 4-year survival of patients post-CRT.

Results:

• 3D-CAE model effectively learned to extract latent features and reconstructed displacements with EPE of 0.0914.
• The survival network had the average AUC value for the ROC curves of 0.76 ± 0.04

Impact: This study used important features in myocardial displacement fields. It gives better AUC in comparison with human-derived descriptors of cardiac motion. Also, there are other parameters that can be added to this model to get a promising 4-year survival prediction.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Primary lower extremity lymphedema

Motivation: Traditional imaging-based staging methods for PLEL rely on subjective assessments by medical professionals and often struggle to capture the micro-level changes associated with lymphedema, which limits the accuracy and granularity of staging.

Goal(s): To explore the value of radiomicss features based on different components extracted from MRI for assessing the staging of PLEL.

Approach: We proposed a machine learning model to integrate multi-region radiomics for automated PLEL staging and employed deep learning for automated subcutaneous tissue segmentation in the lower extremity MRI.

Results: The Dice coefficient for subcutaneous tissue segmentation scored 0.92, and the AUC for lymphedema staging was 0.821.

Impact: The machine learning model based on radiomics in this study shows promising potential and value in lymphedema staging, which is expected to reduce subjective variability and potentially eliminate the need for clinical assays, thus enhancing its clinical applicability.

Keywords: Diagnosis/Prediction, Radiomics, cardiac magnetic resonance

Motivation: The complexity of differentiation between acute and chronic myocardial infarctions(MI) not only complicates the choice of treatment plans but also poses challenges for post-treatment follow-up.

Goal(s): This study aimed to determine whether radiomics model based on T1 mapping can be applied in differential diagnosis of acute and chronic MI.

Approach: Images of 61 patients were included for feature extraction, and the statistically significant features were selected to establish the radiomics model.

Results: This radiomics model demonstrates significant efficacy in distinguishing between acute and chronic myocardial infarction lesions in patients, providing valuable support for clinical diagnosis and follow-up treatment.

Impact: The development of radiomics model to differentiate acute from chronic myocardial infarction lesions holds promise for facilitating prompt clinical decision-making. This advancement enables early medical intervention for MI patients, reducing the risk of adverse cardiovascular events and enhancing patient prognosis.   

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence

Motivation: Manual segmentation of cardiac MRI images is a time-consuming and laborious task prone to observer bias. Automatic segmentation approaches provide poor results in extreme slices. A slice classification step applied before automatic segmentation will lead to better results and reduced variability.

Goal(s): To develop a classifier model with high classification performance on short-axis(SA) cine MRI images for slice selection.

Approach: We trained and compared 2 CNN models for classifying SA cine MRI images into Apical-to-Basal vs Extreme slices.

Results: Xception model had better classification accuracy (0.90) and F1- score (0.93) when compared to InceptionV3 (0.87 and 0.89, respectively).

Impact: The proposed model will provide automatic, fast and accurate classification of MRI cine images, which will improve the accuracy of automatic segmentation of myocardium and its assessment.

Keywords: Diagnosis/Prediction, Joints, meniscus; MRI signal; deep learning

Motivation: The pathogenesis of knee osteoarthritis is affected by many factors.  Among them, meniscus of patients with different stages of disease in magnetic resonance imaging reveal different degrees of abnormal signals.

Goal(s): In this work, abnormal meniscus signals in 400 MRI images of 40 cases were quantified.

Approach: The quantitative indicators were evaluated according to the clinical manifestations of patients.

Results: Results showed that the average area ratios of abnormal meniscal signals were different between the case group and the control group, and the different degrees of abnormal signals could be used as biomarkers of knee osteoarthritis.

Impact: The average area ratios of abnormal signals of meniscus can be used as new biomarkers to provide some objective and accurate biomarkers for knee osteoarthritis.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Osteoarthritis

Motivation: The combination of deep learning, MRI, and clinical data for predicting the time to total knee replacement surgery in knee osteoarthritis patients has been investigated.

Goal(s): The 3D Resnet18 model was employed to extract features from MRI scans, and relevant clinical variables were integrated to establish a comprehensive predictive model. 

Approach: Time-to-surgery probabilities were estimated using the ensemble random survival forest model. The model’s performance was evaluated across clinical variables, two MRI sequences, and their combinations. 

Results: The proposed approach aims to help the precision of TKR surgery decision-making using artificial intelligence.

Impact: This study fuses deep learning, survival analysis, MRI, and clinical data to accurately predict time-to-TKR surgery. Our approach has the potential to enhance TKR surgery decision precision.

Keywords: Diagnosis/Prediction, Cardiomyopathy, Cine, LGE, Hypertrophic cardiomyopathy, RegGAN, CBAM

Motivation: LGE CMR, the standard clinical non-invasive characterization, is limited by its reliance on intravenous contrast agents and long waiting time. Therefore, developing a contrast agent-free technology is essential for achieving fast and cost-effective CMR scans.

Goal(s): To evaluate the reproducibility and reliability of the virtual LGE based on Cine, and compare it with native LGE to assess its efficiency in diagnosing HCM in clinical context.

Approach: RegGAN was employed to forecast LGE imaging and rectify the outcomes. And CBAM was utilized to quantify the influence of diverse components in LGE.

Results: RegGAN-CBAM demonstrates favorable performance in both image and enhancement prediction of LGE.

Impact: The performance of virtual LGE based on Cine exhibits a notable level of diagnostic efficiency and reliability. This approach serves as a non-invasive myocardial tissue characterization method with practical applicability in the clinical assessment of HCM.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Regression, CNN, Radiomics

Motivation: While Radiomics analysis has shown predictive power for plasma cell infiltration (PCI) from MRI in Myeloma patients, convolutional neural networks (CNNs) offer an opportunity for improved performance and generalizability.

Goal(s): Our objective was to develop a predictive model for PCI using CNNs while addressing the challenges posed by limited dataset size.

Approach: CNNs were trained on MRI data of the pelvic bone marrow and its predictive capabilities were enriched by concatenating radiomic features in the latent space.

Results: The findings revealed limitations due to the small dataset size. However, incorporating radiomic features enhanced prediction accuracy, aligning with radiomics and random forest-based methods.

Impact: This study highlights the limitations of deep learning when using a small dataset. It underlines the importance of feature extraction and the need of dedicating substantial efforts to create large annotated datasets.

Keywords: Other AI/ML, Machine Learning/Artificial Intelligence, Radiology Protocols, Decision Support System, Modeling, All-Body MR Protocols

Motivation: We developed a system that performs radiology protocol selection for incoming MRI orders.

Goal(s): To enhance MRI protocol selection accuracy and efficiency. We evaluated new models and expanded anatomic/subspeciality coverage compared to a prior body MRI protocol selection system.

Approach: A machine learning-driven decision-support system was developed integrating kernel-based, tree-based, boosting, and deep-learning algorithms with an ensemble classifier in 22,524 patients. This system utilizes electronic medical records to predict the top-three likely MRI protocols and their probabilities.

Results: A cumulative F1-score of 97.1% for the top-three predicted MRI protocols was obtained in a test set of 3,379 patients.

Impact: The proposed system has the potential to improve radiologists’ protocol selection accuracy by notifying them of protocol-case discrepancies due to the individual patient’s conditions, and to enable a decision-support system for greater efficiency in selecting commonly utilized MR protocols.