Julius Akesson
Postdoctoral Scholar, Radiological Sciences Laboratory
All Publications
-
Rapid online deep artifact suppression for real-time spiral bSSFP CMR with blipped-CAIPI simultaneous multi-slice imaging at 1.5 T.
Magnetic resonance imaging
2026; 134: 110751
Abstract
Real-time (RT) bSSFP MRI enables fast free-breathing cardiovascular imaging but requires 10-16 slices for functional assessment, resulting in prolonged scan times. Simultaneous multi-slice (SMS) imaging can reduce acquisition time but when combined with non-Cartesian trajectories, it relies on iterative reconstructions that often preclude online use. This study investigates deep artifact suppression to facilitate rapid, online reconstruction of RT-SMS.A spiral bSSFP RT sequence with SMS-2 was implemented at 1.5 T. Reconstruction used slice separation in k-space, followed by deep artifact suppression in image space using a 3D U-Net. Ten healthy volunteers were prospectively imaged. RT-SMS image quality and reconstruction time were compared between deep artifact suppression and compressed sensing (CS) reconstructions. Left (LV) and right (RV) ventricular volumes at end-diastole (EDV) and end-systole (ESV) and LV mass (LVM) were compared between RT-SMS with deep artifact suppression and reference-standard breath-hold (BH) imaging.The RT-SMS technique acquired each SMS slice-pair within a single cardiac cycle, reducing the total acquisition by ∼13× compared with BH imaging (15 s vs 3 min 15 s). Total reconstruction for RT-SMS using deep artifact suppression took ∼30 s, compared with 24 min 55 s for compressed sensing. Deep artifact suppression consistently outperformed CS in quantitative and qualitative image quality (p < 0.001). Functional agreement between BH and RT-SMS with deep artifact suppression was good (LVEDV: -7.5 ± 6.8 ml, LVESV: -0.9 ± 4.2 ml, RVEDV: -6.4 ± 8.4 ml, RVESV: 0.2 ± 10.7 ml, LVM: -10.3 ± 11.0 g).Online deep artifact suppression reconstruction for RT-SMS bSSFP CMR enables free-breathing short-axis coverage with a substantial reduction in acquisition and reconstruction time while maintaining diagnostic image quality.
View details for DOI 10.1016/j.mri.2026.110751
View details for PubMedID 42521000
-
Retrospectively synchronized time-resolved ventricular cine images from 2D real-time exercise cardiac magnetic resonance imaging
CLINICAL PHYSIOLOGY AND FUNCTIONAL IMAGING
2025; 45 (5): e70027
Abstract
Breath-hold ECG-gated cardiovascular magnetic resonance (CMR) imaging is challenging during exercise due to motion, ECG-problems, and lengthy scans. To facilitate time-resolved volumetric measures from exercise-CMR, we aimed to develop a method for constructing time-resolved ventricular cines from real-time free-breathing exercise-CMR. Time-resolved ventricular cines were semi-automatically constructed from real-time exercise-CMR by identifying end-expiratory timeframes, identifying one R-R interval within these timeframes, and synchronizing R-R intervals across slice positions. To investigate utility, ECG-gated rest CMR and real-time exercise-CMR images were collected from ten healthy volunteers and ten heart failure patients. The consistency of the left ventricular mass (LVM) was assessed between rest and exercise at end diastole (ED), mid systole (MS), end systole (ES), and early rapid filling (ERF). When comparing LVM between rest and exercise for healthy volunteers, bias ± SD was 1.5 ± 2.7 g at ED, 0.9 ± 3.3 g at MS, 1.3 ± 3.3 g at ES, and 1.2 ± 3.3 g at ERF. When comparing LVM between rest and exercise for heart failure patients, bias ± SD was 1.6 ± 2.8 g at ED, 1.0 ± 2.7 g at MS, 1.5 ± 2.6 g at ES, and 1.6 ± 2.5 g at ERF. The bias ± SD between ED and ES in standard rest images was 0.0 ± 0.7 g for healthy volunteers, and 0.0 ± 0.5 g for heart failure patients. The method for constructing time-resolved ventricular cines from real-time exercise-CMR demonstrated utility for time-resolved volumetric measurements in healthy volunteers and heart failure patients.
View details for DOI 10.1111/cpf.70027
View details for Web of Science ID 001585623700003
View details for PubMedID 40899453
View details for PubMedCentralID PMC12406293
-
Random effects during training: Implications for deep learning-based medical image segmentation.
Computers in biology and medicine
2024; 180: 108944
Abstract
BACKGROUND: A single learning algorithm can produce deep learning-based image segmentation models that vary in performance purely due to random effects during training. This study assessed the effect of these random performance fluctuations on the reliability of standard methods of comparing segmentation models.METHODS: The influence of random effects during training was assessed by running a single learning algorithm (nnU-Net) with 50 different random seeds for three multiclass 3D medical image segmentation problems, including brain tumour, hippocampus, and cardiac segmentation. Recent literature was sampled to find the most common methods for estimating and comparing the performance of deep learning segmentation models. Based on this, segmentation performance was assessed using both hold-out validation and 5-fold cross-validation and the statistical significance of performance differences was measured using the Paired t-test and the Wilcoxon signed rank test on Dice scores.RESULTS: For the different segmentation problems, the seed producing the highest mean Dice score statistically significantly outperformed between 0 % and 76 % of the remaining seeds when estimating performance using hold-out validation, and between 10 % and 38 % when estimating performance using 5-fold cross-validation.CONCLUSION: Random effects during training can cause high rates of statistically-significant performance differences between segmentation models from the same learning algorithm. Whilst statistical testing is widely used in contemporary literature, our results indicate that a statistically-significant difference in segmentation performance is a weak and unreliable indicator of a true performance difference between two learning algorithms.
View details for DOI 10.1016/j.compbiomed.2024.108944
View details for PubMedID 39096609
-
Training deep learning based dynamic MR image reconstruction using open-source natural videos
SCIENTIFIC REPORTS
2024; 14 (1): 11774
Abstract
To develop and assess a deep learning (DL) pipeline to learn dynamic MR image reconstruction from publicly available natural videos (Inter4K). Learning was performed for a range of DL architectures (VarNet, 3D UNet, FastDVDNet) and corresponding sampling patterns (Cartesian, radial, spiral) either from true multi-coil cardiac MR data (N = 692) or from synthetic MR data simulated from Inter4K natural videos (N = 588). Real-time undersampled dynamic MR images were reconstructed using DL networks trained with cardiac data and natural videos, and compressed sensing (CS). Differences were assessed in simulations (N = 104 datasets) in terms of MSE, PSNR, and SSIM and prospectively for cardiac cine (short axis, four chambers, N = 20) and speech cine (N = 10) data in terms of subjective image quality ranking, SNR and Edge sharpness. Friedman Chi Square tests with post-hoc Nemenyi analysis were performed to assess statistical significance. In simulated data, DL networks trained with cardiac data outperformed DL networks trained with natural videos, both of which outperformed CS (p < 0.05). However, in prospective experiments DL reconstructions using both training datasets were ranked similarly (and higher than CS) and presented no statistical differences in SNR and Edge Sharpness for most conditions.The developed pipeline enabled learning dynamic MR reconstruction from natural videos preserving DL reconstruction advantages such as high quality fast and ultra-fast reconstructions while overcoming some limitations (data scarcity or sharing). The natural video dataset, code and pre-trained networks are made readily available on github.
View details for DOI 10.1038/s41598-024-62294-7
View details for Web of Science ID 001230489600031
View details for PubMedID 38783018
View details for PubMedCentralID PMC11116488
-
Diagnostic confidence with quantitative cardiovascular magnetic resonance perfusion mapping increases with increased coverage of the left ventricle
JOURNAL OF CARDIOVASCULAR MAGNETIC RESONANCE
2024; 26 (1): 101007
Abstract
Quantitative cardiovascular magnetic resonance (CMR) first pass perfusion maps are conventionally acquired with 3 short-axis (SAX) views (basal, mid, and apical) in every heartbeat (3SAX/1RR). Thus, a significant part of the left ventricle (LV) myocardium, including the apex, is not covered. The aims of this study were 1) to investigate if perfusion maps acquired with 3 short-axis views sampled every other RR-interval (2RR) yield comparable quantitative measures of myocardial perfusion (MP) as 1RR and 2) to assess if acquiring 3 additional perfusion views (i.e., total of 6) every other RR-interval (2RR) increases diagnostic confidence.In 287 patients with suspected ischemic heart disease stress and rest MP were performed on clinical indication on a 1.5T MR scanner. Eighty-three patients were examined by acquiring 3 short-axis perfusion maps with 1RR sampling (3SAX/1RR); for which also 2RR maps were reconstructed. Additionally, in 103 patients 3 short-axis and 3 long-axis (LAX; 2-, 3, and 4-chamber view) perfusion maps were acquired using 2RR sampling (3SAX + 3LAX/2RR) and in 101 patients 6 short-axis perfusion maps using 2RR sampling (6SAX/2RR) were acquired. The diagnostic confidence for ruling in or out stress-induced ischemia was scored according to a Likert scale (certain ischemia [2 points], probably ischemia [1 point], uncertain [0 points], probably no ischemia [1 point], certain no ischemia [2 points]).There was a strong correlation (R = 0.99) between 3SAX/1RR and 3SAX/2RR for global MP (mL/min/g). The diagnostic confidence score increased significantly when the number of perfusion views was increased from 3 to 6 (1.24 ± 0.68 vs 1.54 ± 0.64, p < 0.001 with similar increase for 3SAX+3LAX/2RR (1.29 ± 0.68 vs 1.55 ± 0.65, p < 0.001) and for 6SAX/2RR (1.19 ± 0.69 vs 1.53 ± 0.63, p < 0.001).Quantitative perfusion mapping with 2RR sampling of data yields comparable perfusion values as 1RR sampling, allowing for the acquisition of additional views within the same perfusion scan. The diagnostic confidence for stress-induced ischemia increases when adding 3 additional views, short- or long axes, to the conventional 3 short-axis views. Thus, future development and clinical implementation of quantitative CMR perfusion should aim at increasing the LV coverage from the current standard using 3 short-axis views.
View details for DOI 10.1016/j.jocmr.2024.101007
View details for Web of Science ID 001289704300001
View details for PubMedID 38316344
View details for PubMedCentralID PMC11211224
-
Deep learning can yield clinically useful right ventricular segmentations faster than fully manual analysis
SCIENTIFIC REPORTS
2023; 13 (1): 1216
Abstract
Right ventricular (RV) volumes are commonly obtained through time-consuming manual delineations of cardiac magnetic resonance (CMR) images. Deep learning-based methods can generate RV delineations, but few studies have assessed their ability to accelerate clinical practice. Therefore, we aimed to develop a clinical pipeline for deep learning-based RV delineations and validate its ability to reduce the manual delineation time. Quality-controlled delineations in short-axis CMR scans from 1114 subjects were used for development. Time reduction was assessed by two observers using 50 additional clinical scans. Automated delineations were subjectively rated as (A) sufficient for clinical use, or as needing (B) minor or (C) major corrections. Times were measured for manual corrections of delineations rated as B or C, and for fully manual delineations on all 50 scans. Fifty-eight % of automated delineations were rated as A, 42% as B, and none as C. The average time was 6 min for a fully manual delineation, 2 s for an automated delineation, and 2 min for a minor correction, yielding a time reduction of 87%. The deep learning-based pipeline could substantially reduce the time needed to manually obtain clinically applicable delineations, indicating ability to yield right ventricular assessments faster than fully manual analysis in clinical practice. However, these results may not generalize to clinics using other RV delineation guidelines.
View details for DOI 10.1038/s41598-023-28348-y
View details for Web of Science ID 000955774100008
View details for PubMedID 36681759
View details for PubMedCentralID PMC9867728
-
Ventricular longitudinal function by cardiovascular magnetic resonance predicts cardiovascular morbidity in HFrEF patients
ESC HEART FAILURE
2022; 9 (4): 2313-2324
Abstract
Ventricular longitudinal function measured as basal-apical atrioventricular plane displacement (AVPD) or global longitudinal strain (GLS) is a potent predictor of mortality and could potentially be a predictor of heart failure-associated morbidity. We hypothesized that low AVPD and GLS are associated with the combined endpoint of cardiovascular mortality and heart failure-associated morbidity.Two hundred eighty-seven patients (age 62 ± 12 years, 78% male) with heart failure with reduced (≤40%) ejection fraction (HFrEF) referred to a cardiovascular magnetic resonance exam were included. Ventricular longitudinal function, ventricular volume, and myocardial fibrosis or infarction were analysed from cine and late gadolinium enhancement images. National registries provided data on causes of cardiovascular hospitalizations and cardiovascular mortality for the combined endpoint. Time-to-event analysis capable of including reoccurring events was employed with a 5-year follow-up. HFrEF patients had EF 26.5 ± 8.0%, AVPD 7.8 ± 2.4 mm, and GLS -7.5 ± 3.0%. In contrast, ventricular longitudinal function was approximately twice as large in an age-matched control group (AVPD 15.3 ± 1.6 mm; GLS -20.6 ± 2.0%; P < 0.001 for both). There were 578 events in total, and the majority were HF hospitalizations (n = 418). Other major events were revascularizations (n = 64), cardiovascular deaths (n = 40), and myocardial infarctions (n = 21). One hundred fifty-five (54%) patients experienced at least one event (mean 2.0, range 0-64). Of these patients, 119 (71%) had three events or fewer, and the first three events comprised 51% of all events (295 events). Patients in the bottom AVPD or GLS tertile (<6.8 mm or >-6.1%) overall experienced more than 3 times as many events as the top tertile (>8.8 mm or <-8.4%; P < 0.001). Patients in this tertile also faced more cardiovascular deaths (P < 0.05), HF hospitalizations (P = 0.001), myocardial infarctions (only GLS: P = 0.032), and accumulated longer in-hospital length-of-stay overall (AVPD 20.9 vs. 9.1 days; GLS 22.4 vs. 6.5 days; P = 0.001 for both), and from HF hospitalizations (AVPD 19.3 vs. 8.3 days; GLS 19.3 vs. 5.4 days; P = 0.001 for both). In multivariate analysis adjusted for significant covariates, AVPD and GLS remained independent predictors of events (hazard ratio 1.12 per-mm-decrease and 1.13 per-%-increase) alongside hyponatremia (<135 mmol/L), aetiology of HF, and LV end-diastolic volume index.Low ventricular longitudinal function is associated with an increase in number of events as well as longer in-hospital stay from cardiovascular causes. In addition, AVPD and GLS have independent prognostic value for cardiovascular mortality and morbidity in HFrEF patients.
View details for DOI 10.1002/ehf2.13916
View details for Web of Science ID 000780981500001
View details for PubMedID 35411699
View details for PubMedCentralID PMC9288769
https://orcid.org/0000-0003-3480-5205