Deep Learning Methods for Abnormality Detection and Segmentation in Computed Tomography and Magnetic Resonance Images

To examine the clinicopathological features and identify the predictors of pathological upgrading in low-risk prostate cancer (PCa) patients without hypointense lesions on the apparent diffusion coefficient (ADC) map calculated from multiparametric magnetic resonance imaging. We reviewed the medical records of 1905 PCa patients who underwent radical prostatectomy between 2007 and 2015. All ADC images were graded using the five-grade Likert scale; the positive hypointense lesions were graded 4-5. We analyzed 256 patients with low-risk classifications according to D'Amico criteria. Patients were classified into two groups according to the pathologic upgrading in the surgical specimens. The predictive factors for pathologic upgrading were evaluated using amultivariate logistic regression analysis. In 256 patients with low-risk PCa, the percentage of positive cores [odds ratio (OR) 1.09; 95% confidence interval (CI) 1.02-1.16], the percentage of cancer in the positive cores (OR 1.07, 95% CI 1.03-1.12), and the presence of hypointensity on an ADC map (OR 2.28; 95% CI 1.23-4.22) were independent predictors of pathologic upgrading. Notably, 138 of low-risk patients (53.9%) had no hypointense lesions on an ADC map. Of these 138 patients, the percentage of positive cores (OR 1.09; 95% CI 1.01-1.18) and the percentage of cancer in the positive cores (OR 1.06; 95% CI 1.01-1.12) remained independent predictors of pathologic upgrading. In low-risk PCa patients without hypointense lesions on an ADC map, biopsy-related parameters such as the percentage of positive cores and the percentage of cancer in the positive cores were independent predictors of pathological upgrading following radical prostatectomy.

This study aims for targeted biopsy validation of magnetic resonance fingerprinting (MRF) and diffusion mapping for characterizing peripheral zone (PZ) prostate cancer and noncancers. One hundred four PZ lesions in 85 patients who underwent magnetic resonance imaging were retrospectively analyzed with apparent diffusion coefficient (ADC) mapping, MRF, and targeted biopsy (cognitive or in-gantry). A radiologist blinded to pathology drew regions of interest on targeted lesions and visually normal peripheral zone on MRF and ADC maps. Mean T1, T2, and ADC were analyzed using linear mixed models. Generalized estimating equations logistic regression analyses were used to evaluate T1 and T2 relaxometry combined with ADC in differentiating pathologic groups. Targeted biopsy revealed 63 cancers (low-grade cancer/Gleason score 6 = 10, clinically significant cancer/Gleason score ≥7 = 53), 15 prostatitis, and 26 negative biopsies. Prostate cancer T1, T2, and ADC (mean ± SD, 1660 ± 270 milliseconds, 56 ± 20 milliseconds, 0.70 × 10 ± 0.24 × 10 mm/s) were significantly lower than prostatitis (mean ± SD, 1730 ± 350 milliseconds, 77 ± 36 milliseconds, 1.00 × 10 ± 0.30 × 10 mm/s) and negative biopsies (mean ± SD, 1810 ± 250 milliseconds, 71 ± 37 milliseconds, 1.00 × 10 ± 0.33 × 10 mm/s). For cancer versus prostatitis, ADC was sensitive and T2 specific with comparable area under curve (AUC; (AUCT2 = 0.71, AUCADC = 0.79, difference between AUCs not significant P = 0.37). T1 + ADC (AUCT1 + ADC = 0.83) provided the best separation between cancer and negative biopsies. Low-grade cancer T2 and ADC (mean ± SD, 75 ± 29 milliseconds, 0.96 × 10 ± 0.34 × 10 mm/s) were significantly higher than clinically significant cancers (mean ± SD, 52 ± 16 milliseconds, 0.65 ± 0.18 × 10 mm/s), and T2 + ADC (AUCT2 + ADC = 0.91) provided the best separation. T1 and T2 relaxometry combined with ADC mapping may be useful for quantitative characterization of prostate cancer grades and differentiating cancer from noncancers for PZ lesions seen on T2-weighted images.

Prostate cancer (PCa) is a highly heterogeneous disease, making tailored treatment approaches challenging. Magnetic resonance imaging (MRI), notably diffusion-weighted imaging (DWI) and the derived Apparent Diffusion Coefficient (ADC) maps, plays a crucial role in PCa characterization. In this context, radiomics is a very promising approach able to disclose insights from MRI data. However, the sensitivity of radiomic features to MRI settings, encompassing DWI protocols and multicenter variations, requires the development of robust and generalizable models. To develop a comprehensive radiomics framework for noninvasive PCa characterization using ADC maps, focusing on identifying reliable imaging biomarkers against intra- and inter-institution variations. Two patient cohorts, including an internal cohort (118 PCa patients) used for both training (75%) and hold-out testing (25%), and an external cohort (50 PCa patients) for independent testing, were employed in the study. DWI images were acquired with three different DWI protocols on two different MRI scanners: two DWI protocols acquired on a 1.5-T scanner for the internal cohort, and one DWI protocol acquired on a 3-T scanner for the external cohort. One hundred and seven radiomics features (i.e., shape, first order, texture) were extracted from ADC maps of the whole prostate gland. To address variations in DWI protocols and multicenter variability, a dedicated pipeline, including two-way ANOVA, sequential-feature-selection (SFS), and ComBat features harmonization was implemented. Mann-Whitney U-tests (α=0.05) were performed to find statistically significant features dividing patients with different tumor characteristics in terms of Gleason score (GS) and T-stage. Support-Vector-Machine models were then developed to predict GS and T-stage, and the performance was assessed through the area under the curve (AUC) of receiver-operating-characteristic curves. Downstream of ANOVA, two subsets of 38 and 41 features stable against DWI protocol were identified for GS and T-stage, respectively. Among these, SFS revealed the most predictive features, yielding an AUC of 0.75 (GS) and 0.70 (T-stage) in the hold-out test. Employing ComBat harmonization improved the external-test performance of the GS model, raising AUC from 0.72 to 0.78. By incorporating stable features with a harmonization procedure and validating the model on an external dataset, model robustness, and generalizability were assessed, highlighting the potential of ADC and radiomics for PCa characterization.

This study aimed to evaluate the effectiveness of apparent diffusion coefficient (ADC) parameters in distinguishing between Prostate Cancer Research International Active Surveillance (PRIAS) non-reclassification and reclassification groups during active surveillance (AS) of prostate cancer. We included 55 patients who fulfilled the PRIAS criteria and underwent ≥ 2 magnetic resonance imaging (MRI) including diffusion-weighted imaging with an interval of ≤ 3years between baseline and second MRI. A mono-exponential fitting model was used to automatically create ADC maps with minimum b-values of 0 and maximum of 2000s/mm2. For detectable lesions on ADC maps, the lesions were manually segmented on each slice of the ADC maps. For undetectable lesions, the corresponding normal-appearing zone of the lobe on each slice of ADC maps was segmented. The ADC data for each slice were summed to obtain the 25th, 50th, and 75th percentile ADC values of the histogram at baseline and second MRI. These ADC parameters at baseline and second MRI, and the changes of ADC parameters from baseline to second MRI were compared between PRIAS non-reclassification and reclassification groups. The PRIAS reclassification group had significantly lower 25th, 50th, and 75th percentile ADC values at second MRI compared to the non-reclassification group. The non-reclassification group had significantly lower changes in ADC values in these percentiles compared to the reclassification group. The ADC parameters at second MRI and the changes from baseline to second MRI may be effective distinguishing factors between PRIAS non-reclassification and reclassification groups.

The aim of this study was to evaluate an advanced magnetic resonance imaging (MRI) method, apparent diffusion coefficient (ADC) mapping, in the functional assessment of carious teeth. 38 extracted human teeth with scores of 0, 3 and 6 according to International Caries Detection and Assessment System (ICDAS) criteria were screened and subsequently analyzed by MRI at 2.35 T. Histology sectioning of teeth was used for the gold standard by analyzing two extreme cases (intact and severely decayed). ADC maps of the same teeth were calculated from corresponding diffusion-weighted images and used to obtain ADC distributions along dental pulp as functions of the relative pulp length measured from the occlusal pulp side. The measured distributions were analyzed for the best fit by a four-parameter three-segment linear regression model for ADC distribution along the pulp. MRI results were in good agreement with findings in histological sections of identical teeth. The best fit model parameters, relative decayed region depth, relative transition region width and ADC values of intact and decayed pulp tissue, showed statistically significant differences between the ADC values of intact and decayed pulp tissue (1.0 × 10^-9 m²/s vs. 0.74-0.89 × 10^-9 m²/s) and the relative decayed region depth progressing with ICDAS score (3 vs. 46% with ICDAS 3 vs. ICDAS 6). The results of this feasibility study confirmed relevance of ADC mapping for the discrimination and localization of intact and decayed regions in dental pulps of carious teeth.

The objective of our study was to investigate the comparative effectiveness of different MRI sequences for the estimation of index lesion volume in patients with prostate cancer (PCa) compared with ground truth volume measured on whole-mount pathology. Patients with PCa underwent multiparametric MRI (mpMRI) on a 3-T MRI scanner before radical prostatectomy. Forty PCa index lesions were identified and outlined on histology by a pathologist. Two radiologists who were informed about the presence of PCa but were not aware of lesion outlines on histology worked in consensus to delineate PCa lesions on T2-weighted imaging, apparent diffusion coefficient (ADC) maps, and early-phase dynamic contrast-enhanced MRI (DCE-MRI). The lesion volumes from different mpMRI sequences and the percentage of volume underestimation compared with pathology were calculated and correlated with volume at pathology. The repeated-measures ANOVA with the posthoc Bonferroni test was performed to evaluate whether the difference between the estimated tumor volumes was statistically significant. The mean PCa lesion volume estimated from pathology, T2-weighted imaging, DWI (ADC maps), and DCE-MRI were 4.61 ± 4.99 (SD) cm3, 2.03 ± 2.96 cm3, 1.81 ± 2.76 cm3, and 3.48 ± 4.06 cm3, respectively. The lesion volumes on T2-weighted images (p = 0.000002), ADC maps (p = 0.000003), and DCE-MR images (p = 0.004412) were significantly lower than those from pathology. PCa lesion volume was significantly underestimated on T2-weighted images, ADC maps, and DCE-MR images compared with pathology by 54.98% ± 22.60% (mean ± SD), 58.59% ± 18.58%, and 18.33% ± 30.11%, respectively; underestimation using T2-weighted imaging (p = 1.01 × 10-11) and DWI (p = 2.94 × 10-11) was significantly higher than underestimation using DCE-MRI. Correlations between lesion volume estimated on T2-weighted images, ADC maps, and DCE-MR images with pathology were 0.91 (p = 9.03 × 10-16), 0.86 (p = 7.32 × 10-13), and 0.93 (p = 8.22 × 10-18), respectively. DCE-MRI performed better than T2-weighted imaging and DWI for estimation of index PCa volume and therefore can be preferred over these other two sequences for volume estimation.

Malignant peripheral nerve sheath tumors (MPNSTs) are rare neoplasms requiring differentiation from benign neurogenic tumors on diagnostic imaging. We evaluated the diagnostic utility of texture analysis (TA) in distinguishing seven pathologically confirmed MPNSTs from eight schwannomas using magnetic resonance imaging (MRI), including T1- (T1WI) and T2-weighted imaging (T2WI), and apparent diffusion coefficient (ADC) maps. TA's performance was compared with that of conventional approaches using ADC values alone. Tumors were segmented, and 90 texture features were extracted using LIFEx software. Significantly different features (p < 0.05) were identified using the Mann-Whitney U test and evaluated through receiver operating characteristic (ROC) analysis. ADC maps were used to measure the minimum, mean, and maximum ADC values, followed by ROC analysis. Two T2WI-based texture features (neighborhood grey-tone difference matrix Contrast and grey-level size zone matrix Grey Level Variance) demonstrated the highest diagnostic performance (area under the curve = 0.911 [95% confidence interval (CI):0.755-1.000]), comparable to the minimum and mean ADC values (area under the curve = 0.898 [95% CI:0.691-1.000]). TA may help differentiate MPNSTs from schwannomas. T2WI-based TA offers a viable alternative when ADC maps are limited by magnetic susceptibility artifacts. J. Med. Invest. 72 : 367-374, August, 2025.

We aimed to identify the radiologic features of uterine cervical adenocarcinoma associated with lobular endocervical glandular hyperplasia (LEGH). We retrospectively analyzed magnetic resonance (MR) images and pathologic findings of eight patients who underwent preoperative MR imaging followed by surgical resection and who were pathologically diagnosed with adenocarcinoma (except for adenocarcinoma in situ) associated with LEGH. We assessed the following MR findings: multicystic component (MC), solid component (SC), signal intensity of SC on diffusion-weighted imaging (DWI) and the apparent diffusion coefficient (ADC) map, and radiological stage (r-stage) based on the FIGO classification. A pathologist reevaluated the pathological stage (p-stage) according to the FIGO classification. We correlated the MR findings with the pathologic features. Eight patients were classified into the following three types based on the MR findings: type A, MC and SC; type B, only SC; and type C, only MC. In the five patients with type A, diffusion restriction (DR) was seen on DWI and the ADC map. In 80% of type A cases, the r-stage matched the p-stage. In the one patient with type B, DR was not seen on DWI or the ADC map, and the r-stage matched the p-stage. In the remaining type C cases, DR was not seen on DWI or the ADC map, and the r-stage was underestimated compared with the p-stage. On MR imaging, the most common type of adenocarcinoma with LEGH is type A; type C is difficult to diagnose as carcinoma.

Background: Osteoporosis is a systemic skeletal disorder marked by reduced bone density and microarchitectural deterioration, leading to increased fracture risk. While the dual-energy X-ray absorptiometry (DEXA) scan is the World Health Organization (WHO)-recommended diagnostic standard, its limitations necessitate alternative methods. Emerging magnetic resonance imaging (MRI) techniques, radiomics, and machine learning promise to enhance osteoporosis diagnosis through detailed analysis of lumbar MRI apparent diffusion coefficient (ADC) maps, potentially revolutionizing early detection and treatment strategies. Objectives: In this study, we are going to evaluate the performance of machine learning (ML) models using radiomics features of lumbar MRI ADC map for osteoporosis detection, and to identify significant features and their diagnostic thresholds. Specific performance metrics such as accuracy, sensitivity, specificity, and Area Under the receiver operating characteristic (ROC) Curve (AUC) were assessed. Patients and Methods: This retrospective study employed a cross-sectional design, with a total of 140 cases, including 21 with osteoporosis. The study's inclusion criteria consisted of concurrent lumbar MRI and DEXA within a year, while exclusion criteria included infectious or neoplastic lumbar lesions, fractures, instrumentation, significant osteodegenerative changes, cases where the first four lumbar vertebrae were not included in the imaging field, and absence of diffusion-weighted imaging. Manual segmentation of lumbar vertebrae from ADC maps was performed to create a comprehensive dataset, comprising 5,580 radiomics features per case. Subsequently, the top five features selected by fast correlation-based filter (FCBF) were used to test the performance of seven Machine Learning algorithms (k-Nearest neighbors, decision tree, random forest, logistic regression, support vector machine, naive bayes, and neural network). Statistical tests and ROC curve analysis were conducted to determine the significance and thresholds of these features. Results: The study included 140 cases, with 132 females (94.3%) and 8 males (5.7%), and a mean age of 65.32 ± 8.50 years. The mean BMI was 31.43 ± 5.53 kg/m² for females and 26 ± 3.59 kg/m² for males. In terms of demographic differences, no significant age difference was found between the osteoporotic and non-osteoporotic groups (P = 0.889). However, the osteoporotic group had significantly lower mean body weight (64.90 ± 10.13 kg vs. 74.68 ± 13.94 kg, P = 0.003) and BMI (27.40 ± 4.38 kg/m² vs. 31.77 ± 5.52 kg/m², P = 0.001) compared to the non-osteoporotic group. The median interval between DEXA and lumbar MRI was 1 month (range 0.1 - 11.87 months). The Neural Network model demonstrated the highest performance with an AUC of 0.616 and a classification accuracy of 0.764 using all features. The Naive Bayes model, using the top five features selected by FCBF, showed the highest performance with an AUC of 0.913, accuracy of 0.907, sensitivity of 0.667, and specificity of 0.95. All ML models’ performance were elevated by feature selection. Independent t-tests and Mann-Whitney U tests identified 521 and 670 significant features, respectively (P &lt; 0.05). ROC analysis revealed 58 features with AUC values above 0.70. Conclusion: This study's findings suggest that ML models, particularly the Naive Bayes algorithm, can effectively use lumbar ADC map radiomics to diagnose osteoporosis. These findings could enhance early detection and treatment strategies, potentially improving patient outcomes and reducing the burden of osteoporotic fractures. This study also established threshold values for significant features.

A retrospective evaluation of diffusion-weighted imaging, apparent diffusion coefficient (ADC) maps, and T2-weighted images in patients with cervical compression myelopathy. To correlate high signal intensity on ADC maps and T2-weighted images to neurologic severity and radiologic spinal cord compression. Previous studies indicated that the ADC map was more sensitive in detection of cervical compressive myelopathy than T2-weighted imaging. The relationship to neurologic severity has not been previously published. In 100 patients with or without cervical compressive myelopathy, the magnetic resonance appearance of the spinal cord on ADC maps and T2-weighted images was evaluated independently by 2 readers. On the basis of the presence or absence of abnormality, patients were categorized into 4 groups by type of intramedullary change. The degree of canal compression and the Japanese Orthopedic Association (JOA) score for cervical myelopathy were compared among the 4 types. The type I group consisted of 32 patients without signal change on either ADC maps or T2-weighted images. The type II group had 33 patients with high signal intensity only on the ADC map. The type III group contained 28 patients with high signal intensity on both the ADC map and the T2-weighted images. The type IV group comprised 7 patients with high signal intensity only on T2-weighted images. The degree of canal compression and the JOA scores were significantly different among the 4 groups (P < 0.005). Most type III (25 of 28) and type IV (7 of 7) patients had severe cord compression. Average JOA scores in each type were (I) 16.0, (II) 14.7, (III) 11.7, and (IV) 8.7. ADC maps demonstrated internal changes in the early stages of chronic spinal cord compression, but had limitations for the detection of intramedullary changes in late-stage myelopathy.

To examine possible differences between the evolution of cerebral watershed infarction (WI) and that of territorial thromboembolic infarction (TI) by using diffusion-weighted (DW) and T2-weighted magnetic resonance (MR) images and apparent diffusion coefficient (ADC) maps. Fourteen patients with TI and nine with WI underwent MR imaging from the acute to chronic infarction stages. ADC maps were derived from DW images. Lesion-to-normal tissue signal intensity ratios on ADC maps (rADC), echo-planar T2-weighted images, and DW images were calculated. Lesion volumes at acute or early subacute infarction stages were measured on DW images, and final lesion volumes were estimated on fluid-attenuated inversion-recovery images. Analysis of variance revealed a significant difference in temporal evolution patterns of rADC between WI and TI (P <.001). rADC pseudonormalization following TI began about 10 days after symptom onset, but that following WI did not occur until about 1 month after symptom onset. The Pearson correlation coefficient between final and initial infarct volumes was 0.9899 for both infarction subtypes, indicating that the initial ischemic injury volume measured at the acute or early subacute stage predicted the final lesion volume fairly well. The evolution time of ADC is faster for TI than for WI. This difference, which likely originates from the different pathophysiologic and hemodynamic features of the two infarction types, might account for the relatively large range of ADC values reported for the time course of ischemic strokes.

Early detection of disc degeneration and categorizing them is very important for the treatment plan and better prognosis. We aimed to assess the efficacy of apparent diffusion coefficient (ADC) mapping and T2 mapping in the early detection and grading of the severity of degenerative disc disease. In this review, the articles published until July 2022, from the PubMed database were used. All articles contained information about the application of ADC and T2 mapping in the detection of disc degeneration. A total of 40 articles were included in this review. We assessed each article for the feasibility of both techniques to categorize the degeneration. ADC values of nucleus pulposus (NP) showed a negative correlation with Pfirrmann grades and age. T2 values of NP and annulus fibrosus (AF) are significantly decreased with an increase in Pfirrmann grades. And also, as age increases, T2 values of NP and posterior AF decreased significantly. ADC and T2 mapping techniques are sensitive in the detection and categorization of disc degeneration. It is highly recommended to include quantitative magnetic resonance imaging techniques in routine spine imaging protocol in a patient with complaints of back pain.

BackgroundDespite the established role of late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) in characterizing chronic myocardial infarction (MI), a significant portion of chronic MI patients are contraindicative for the use of contrast agents. One promising alternative contrast free technique is diffusion weighted CMR (dwCMR), which has been shown ex vivo to be sensitive to myocardial fibrosis. We used a recently developed in vivo dwCMR in chronic MI pigs to compare apparent diffusion coefficient (ADC) maps with LGE imaging for infarct characterization.MethodsIn eleven mini pigs, chronic MI was induced by complete occlusion of the left anterior descending artery for 150 minutes. LGE, cine, and dwCMR imaging was performed 8 weeks post MI. ADC maps were derived from three orthogonal diffusion directions (b = 400 s/mm2) and one non-diffusion weighted image. Two semi-automatic infarct classification methods, threshold and full width half max (FWHM), were performed in both LGE and ADC maps. Regional wall motion (RWM) analysis was performed and compared to ADC maps to determine if any observed ADC change was significantly influenced by bulk motion.ResultsADC of chronic MI territories was significantly increased (threshold: 2.4 ± 0.3 μm2/ms, FWHM: 2.4 ± 0.2 μm2/ms) compared to remote myocardium (1.4 ± 0.3 μm2/ms). RWM was significantly reduced (threshold: 1.0 ± 0.4 mm, FWHM: 0.9 ± 0.4 mm) in infarcted regions delineated by ADC compared to remote myocardium (8.3 ± 0.1 mm). ADC-derived infarct volume and location had excellent agreement with LGE. Both LGE and ADC were in complete agreement when identifying transmural infarcts. Additionally, ADC was able to detect LGE-delineated infarcted segments with high sensitivity, specificity, PPV, and NPV. (threshold: 0.88, 0.93, 0.87, and 0.94, FWHM: 0.98, 0.97, 0.93, and 0.99, respectively).ConclusionsIn vivo diffusion weighted CMR has potential as a contrast free alternative for LGE in characterizing chronic MI.

Purpose To compare single-shot echo-planar diffusion-weighted imaging-derived apparent diffusion coefficient (ADC) maps with fat-saturated (FS) proton density (PD)-weighted turbo spin-echo (TSE) imaging in the detection of bone marrow lesions (BMLs) after knee trauma. Materials and Methods Institutional review board approval was obtained from Leipzig University. Written informed consent was waived. Three radiologists retrospectively re-examined 97 consecutive patients with reported knee trauma who underwent 1.5-T magnetic resonance (MR) imaging within 90 days of knee trauma. The following sequences were used: (a) sagittal T1-weighted TSE and FS PD-weighted TSE and (b) sagittal T1-weighted TSE and single-shot echo-planar diffusion-weighted imaging-derived ADC mapping. BMLs on the lateral and medial femoral condyle, lateral and medial aspect of the tibial plateau, and patella were documented. Volumetry was performed on BMLs with a thickness of at least 15 mm (major BMLs). ADC values were measured in intact bone marrow and major BMLs. A McNemar test and t tests were used as appropriate to test for significant differences between BML number and volume at an α level of .05. Results Significantly more patients showed at least one BML on ADC maps (98%, 95 of 97 patients) than on FS PD-weighted TSE images (86%, 84 of 97 patients) (P < .001). Of the affected regions detected on FS PD-weighted TSE images, 97% (170 of 175 regions) were identified consistently on ADC maps. Only 58% of the affected regions detected on ADC maps (170 of 293 regions) were identified on FS PD-weighted TSE images (P < .001). Median volume of concordant major BML was approximately two times larger on ADC maps (81 cm3) than on FS PD-weighted TSE images (39 cm3) (P < .001). The ADC values of intact bone marrow and BMLs did not overlap. Conclusion ADC maps are more sensitive than corresponding FS PD-weighted TSE images for detection of BML after knee trauma and allow detection of significantly more and larger BMLs. ADC map evaluation improves diagnostic performance in regions with insufficient spectral fat saturation, such as the patella. © RSNA, 2016 Online supplemental material is available for this article.

To study the effects of cardiac devices on three-dimensional (3D) late gadolinium enhancement (LGE) MRI and to develop a 3D LGE protocol for implantable cardioverter defibrillator (ICD) patients with reduced image artifacts. The 3D LGE sequence was modified by implementing a wideband inversion pulse, which reduces hyperintensity artifacts, and by increasing bandwidth of the excitation pulse. The modified wideband 3D LGE sequence was tested in phantoms and evaluated in six volunteers and five patients with ICDs. Phantom and in vivo studies results demonstrated extended signal void and ripple artifacts in 3D LGE that were associated with ICDs. The reason for these artifacts was slab profile distortion and the subsequent aliasing in the slice-encoding direction. The modified wideband 3D LGE provided significantly reduced ripple artifacts than 3D LGE with wideband inversion only. Comparison of 3D and 2D LGE images demonstrated improved spatial resolution of the heart using 3D LGE. Increased bandwidth of the inversion and excitation pulses can significantly reduce image artifacts associated with ICDs. Our modified wideband 3D LGE protocol can be readily used for imaging patients with ICDs given appropriate safety guidelines are followed.