Keywords: Other AI/ML, Machine Learning/Artificial Intelligence, Arterial Spin Labeling

Motivation: The low SNR of arterial spin labeling (ASL) perfusion weighted images (PWI) directly affects the accuracy of cerebral blood flow (CBF) quantification.

Goal(s): Our goal was to improve SNR of ASL PWI and access accurate CBF quantification.

Approach: We developed XtwASL, a deep learning model for denoising ASL PWI. Evaluation included SNR, peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) for PWI quality as well as ICC and Bland-Altman analysis for CBF assessment.

Results: Compared with other methods, the proposed XtwASL significantly improved SNR of ASL in the case of short acquisition time and the CBF after XtwASL denoising was more accurate.

Impact: The ASL denoising method XtwASL not only improves SNR but also has good clinical usability in CBF quantification, and this may reduce patient discomfort and artifacts caused by long scan time especially in multi-PLD ASL and high-resolution ASL field.

Keywords: Diagnosis/Prediction, Perfusion, Quality Control

Motivation: Quality Control (QC) of ASL data is primarily a manual and subjective process that is time-intensive and can yield inconsistent results due to rater variability, while existing tools provide limited diagnostic metrics mostly focused on specific error source.

Goal(s): To develop a QC detector to automatically detect outliers for ASL data. 

Approach: VAE-GAN model was applied to extract the latent representation of ASL data by which the decision boundary can be determined.

Results: The AUROC of our QC detector on test dataset is 0.82 with accuracy=0.81.

Impact: Our QC detector could help radiologists and researchers working on ASL MRI to automatically identify outliers. Consequently, appropriate operations can be used to correct or exclude outliers to avoid biases in the outcomes and ensure accurate interpretations.

Keywords: Diagnosis/Prediction, Machine Learning/Artificial Intelligence, Plaque analysis;Vascular centerline extraction

Motivation: Magnetic Resonance Vessel Wall Imaging is a crucial method for assessing arterial plaques.

Goal(s): Achieving a rapid and accurate algorithm for vascular centerline extraction and automatic detection of the pituitary stalk, thereby enabling fast, automatic, and quantitative analysis of plaques.

Approach: Proposed a point detection algorithm based on V-NET, designed to achieve rapid and accurate extraction of vascular centerlines and automatic localization of the pituitary stalk.

Results: Accurate key point detection had been achieved, enhancing the precision of vascular centerline extraction and enabling fast and accurate automatic localization of the pituitary stalk.

Impact: The accuracy of vascular centerline detection has been improved, and rapid and automatic quantitative analysis of plaque enhancement has been realized. This can assist in enhancing diagnostic efficiency in clinical settings.

Keywords: AI Diffusion Models, White Matter, WMH segmentation, Domain Generalization, Normalization

Motivation: Self-adaptation normalization (SAN) method generalizes lesion segmentation to new sites by using a linear generator to convert inputs to a site-independent style. But it fails when the domain difference is mostly nonlinear.

Goal(s): Develop a semi-nonlinear self-adaptation network(SNSAN) to generalize the White Matter Hyperintensity(WMH) segmentation model to an external site.

Approach: To replace the linear generator, the method blends two SAN results with a pseudo correlation map, later use the gradient reversal method to guide the result to a site-unrelated style. 

Results: SNSAN normalizes the input data close to a Gaussian distribution and improves the generalization performance on the data from external site.

Impact: We provide a simple and efficient semi-nonlinear normalization method to enhance the domain generalization, and its  performance is better than SAN when the domain gap is affected by more nonlinear factors.

Keywords: Diagnosis/Prediction, Vessels, Image super-resolution , Generative Adversarial Network (GAN) , Multi- path generator model , Medical imaging

Motivation: The project aims to improve MRI image resolution, addressing the limitations of current super-resolution methods that hinder accurate medical diagnosis and treatment planning.

Goal(s): Enhancing MRI image clarity through ResoNet, an innovative approach featuring a multi-path generator, self-adaptive kernel sizing, and a comprehensive loss function.

Approach: Multi-path generator enhances resolution with different kernel sizes, preserving details. Self-adaptive component optimizes kernel size for efficiency. Comprehensive loss function combines GAN-based discriminator, VGG16, and WGAN-GP for sTab. training and quality samples.

Results: ResoNet significantly enhances SSIM and PSNR scores, promising substantial impact in medical imaging for more precise diagnoses and treatment planning by healthcare professionals.

Impact: This work improves MRI diagnostics by enhancing image resolution, enabling clearer visuals for accurate medical assessments. It optimizes performance and resources, accelerates scanning time, and potentially transforms emergency medical practices, making it an improvement in medical imaging.

Keywords: AI/ML Image Reconstruction, Multi-Contrast

Motivation: The conventional multiple sclerosis(MS) protocol in the brain normally takes from 30 minutes to several hours. It often involves multiple 3D series for the quantitative metrics and reports.

Goal(s): To reduce the scan time for the brain MS protocols with more than one 3D series

Approach: We proposed a framework that utilized a 3D T1 and 2D FLAIR as inputs to generate a 3D T2 FLAIR.

Results: The method generates comparable image resolution on all orientations as the acquired 3D FLAIR, and outperforms other methods. The lesion segmentation masks show high consistency. Furthermore, the method demonstrates robustness in the cases with motion.

Impact: The method provides the solution of generating a volumetric FLAIR using the same time as a 2D series and achieving comparable resolutions and diagnosis accuracy as the 3D FLAIR.

Keywords: Diagnosis/Prediction, Tumor, Hemangioblastoma, Contrastive Learning

Motivation: Brainstem and cerebellar hemangioblastoma is a rare tumor with a high risk of haemorrhage during the biopsy. However, it is still a challenging task to distinguish HB from other types of intracranial tumors solely based on neuroimaging techniques.

Goal(s): To propose a computer-aided diagnosis method to classify hemangioblastoma.

Approach: We propose a patient-level classification framework using multi-task supervised contrastive learning, named LaSCL-PLC, for hemangioblastoma classification.

Results: We evaluated the proposed model on a local MRI dataset of brainstem-and-cerebellum tumors, consisting of 97(HB) and 143(others). The experimental results show that our model achieves competitive performance as neuroradiologists. 

Impact: It could improve the preoperative diagnosis hemangioblastoma, which is crucial for clinical treatments.

Keywords: Analysis/Processing, COVID-19, Long-COVID, Multimodal, MRI, myelin.

Motivation: Some people exposed to SARS-CoV-2 experience persistent symptoms months after recovering from acute infection. These symptoms, grouped under the umbrella of “long-COVID”, include neurological and autonomic dysfunctions for which there is not yet a clear explanation.

Goal(s): To identify biophysically meaningful features characterising long-COVID pathophysiology in the brain.

Approach: A novel multimodal MRI protocol was developed to measure iron accumulation, myelin content, inflammation, microstructure alterations, atrophy, and cerebral blood flow changes. Machine learning was employed to discriminate long-COVID from controls and people who fully recovered.

Results: Alterations in bound-pool fraction (myelin content) emerge in long-COVID.

Impact: This study encourages investigating the neurological aspect of long-COVID and its interplay with conditions affecting myelin as it shows that long-COVID may cause myelin alterations or be more likely to occur in people with compromised myelin content in the brain.

Keywords: Diagnosis/Prediction, Brain

Motivation: Many brain disorders present with asymmetrical features, which can be utilized to advance the classification of disease through deep learning.

Goal(s): We aim to investigate how asymmetry awareness can improve model performance to identify brain disorders, subsequently facilitating better early diagnosis and intervention.

Approach: Our approach involved initial pretraining with a focus on asymmetry awareness using conditional contrastive learning. Subsequently, finetuning was performed for various downstream tasks with limited training data.

Results: Our approach is proven superior in identifying brain disorders compared with baseline methods that either directly trained for specific diseases or with pretrained model without considering symmetrical features of input images.

Impact: Our research pioneers the integration of asymmetry awareness in pretraining models for the detection of brain disorders. The superior performance fully illustrated the feasibility and potential of leveraging the asymmetry nature of brain disease for broad deep learning tasks.  
 
 

Keywords: Diagnosis/Prediction, Brain, Phenotypic prediction, structural MRI, transfer learning

Motivation: Small sample size on structural MRI is evitable in reality and significantly limits phenotypic prediction performance. 

Goal(s): Our goal was to improve prediction performance on small datasets for structural MRI brain imaging. 

Approach: We adapted the meta-matching framework from functional to structural MRI, and compared it with baseline methods (Elastic net and direct transfer learning).

Results: Our meta-matching-based approaches can greatly boost behavioral prediction performance for different small-scale structural MRI datasets.

Impact: Our meta-matching-based methods should be able to make good predictions for a variety of neurological and psychiatric disorders even if the availability of structural MRI brain imaging is quite small. 

Keywords: Diagnosis/Prediction, Diffusion Tensor Imaging

Motivation: Assessing white matter integrity may enhance intracerebral haemorrhage (ICH) surgical trial selection. However, patients often only receive CT imaging, on which delineating white matter is challenging.

Goal(s): This study aimed to train a model to delineate white matter in CT scans, using paired DTI tractography maps, and test if the model predicted outcome in an external dataset.

Approach: Tractography maps were generated from the DTI images. A nnU-Net model was trained on paired CT and registered tractography data, and then run on an external dataset of ICH diagnostic CT scans.

Results: The model performed at 58% Dice, and significantly predicted outcome after ICH.

Impact: Our model can predict tractography labels of the corticospinal tract on diagnostic CT scans without the need for DTI, allowing an enhanced prediction of outcome after intracerebral haemorrhage, and potentially leading to more informed selection of candidates for surgical trials.

Keywords: Analysis/Processing, Susceptibility

Motivation: Familial cerebral cavernous malformation (CCM) cases typically present with multi-focal lesions of varying size, count, and location making automatic segmentation for studying lesion burden challenging.

Goal(s): We aimed to develop a deep learning framework for automated detection and volumetric quantification for ongoing large-scale, multi-site evaluation of patients with CCM.

Approach: We implemented a framework that ensembles two-staged deep learning networks for large and small CCM lesions.

Results: Our model achieved an overall Dice-score of 70%. We demonstrated the feasibility of a deep learning approach for detecting and segmenting various-sized lesions in patients with CCM using SWI acquired from multiple sites with various parameters.

Impact: Our study found that total lesion volume quantified with our deep learning approach was statistically associated with increased odds of intracranial hemorrhage history. This demonstrates the potential benefit of our approach in providing a more accurate assessment of disease burden

Keywords: Analysis/Processing, Machine Learning/Artificial Intelligence, T2-FLAIR, white matter hyperintensity, dementia, cholinergic pathway

Motivation: Cholinergic Pathways Hyperintensities Scale (CHIPS) is a visual rating scale to evaluate the burden of cholinergic white matter hyperintensities in T2-FLAIR image, indicating the severity of dementia. However, it is still time-consuming to screen slices throughout the whole brain to choose 4 specific slices for rating.

Goal(s): To develop a deep-learning-based model to automatically select 4 slices specific to CHIPS.

Approach: We used ADNI T2-FLAIR dataset (N=150) to train a 4-class slice classification model (BSCA) utilized by ResNet, and a local dataset (N=30) to test its performance.

Results: Our model achieved the accuracy of 99.82% and F1-score of 99.83%.

Impact: BSCA can be an automatic screening tool to efficiently provide 4 specific T2-FLAIR slices covering the white matter landmarks along the cholinergic pathways for clinicians to help evaluate whether patients have the high risk to develop clinical dementia.

Keywords: Diagnosis/Prediction, Cancer

Motivation: The project was motivated by the need to reduce reliance on contrast agents in MRI for brain tumor assessment, addressing issues of cost and patient discomfort associated with their use.

Goal(s): To improve the visibility and recognition of tumors in T1-weighted MRI scans, eliminating the need for contrast agents.

Approach: The approach involved registration, data augmentation, Pix2Pix-based image generation, and post-processing.

Results: The results demonstrated that the method significantly improved tumor discernibility in T1-weighted MRI scans, offering a valuable diagnostic tool without the need for contrast agents.

Impact: Our research provides an effective solution to enhance brain tumor visualization in MRI, directly benefiting patients and healthcare providers. By eliminating the need for contrast agents, our approach reduces costs, discomfort, and potential risks, advancing brain tumor diagnostics.

Keywords: Diagnosis/Prediction, Cancer, Analysis/Processing, Robustness

Motivation: Motivated by the challenge of enhancing the robustness of deep neural network decisions against
variable noise in MRI-based brain tumor segmentation.

Goal(s): This study aims to evaluate the efficacy of probabilistic bottlenecks in enhancing segmentation robustness.

Approach: Our approach simulates structured perturbations at increasing strength
to assess their impact on segmentation performance utilizing the Wasserstein distance between
per-sample Dice score distributions and the sensitivity with respect to the perturbation strength.

Results: Results show probabilistic bottlenecks significantly increase robustness to Gaussian noise, yet
offer limited improvement towards Gaussian blur, with varying results for other perturbations,
highlighting the perturbation-specific nature of network resilience.

Impact: This study provides a tool to assess and guard against various perturbations in deep learning.It specifically demonstrates that probabilistic bottlenecks boost robustness of performance withrespect to certain noise types, but not all.

Keywords: Diagnosis/Prediction, Tumor, Post cavity segmentation

Motivation: Currently, manual/semiautomatic methods are being used for post-surgical cavity segmentation in glioma patients, which is crucial for precise treatment planning and monitoring. 

Goal(s): To evaluate the potential of deep learning models for automatic post-surgical cavity segmentation.

Approach: Deep learning models were trained on pre-surgery and post-surgery (within 72hours) MRI data and tested on post-surgery data. The concept of data augmentation and transfer learning were also used. 

Results: Initially trained on the pre-surgery BraTS’2021 challenge dataset, the models underwent further refinement with local hospital's post-surgical data, resulting in 36% improvement in post-surgical cavity segmentation (U-Net₃, DSC = 0.64).

Impact: This study demonstrates the potential of deep-learning in automatic segmentation of post-surgical cavity on MRI images. Proposed automatic segmentation models can assist in fast and objective treatment planning. Further optimizations for improved accuracy and validation on large dataset is required.