skip to main content

Title: Evaluating the relationship between magnetic resonance image quality metrics and deep learning‐based segmentation accuracy of brain tumors
Abstract Background

Magnetic resonance imaging (MRI) scans are known to suffer from a variety of acquisition artifacts as well as equipment‐based variations that impact image appearance and segmentation performance. It is still unclear whether a direct relationship exists between magnetic resonance (MR) image quality metrics (IQMs) (e.g., signal‐to‐noise, contrast‐to‐noise) and segmentation accuracy.


Deep learning (DL) approaches have shown significant promise for automated segmentation of brain tumors on MRI but depend on the quality of input training images. We sought to evaluate the relationship between IQMs of input training images and DL‐based brain tumor segmentation accuracy toward developing more generalizable models for multi‐institutional data.


We trained a 3D DenseNet model on the BraTS 2020 cohorts for segmentation of tumor subregions enhancing tumor (ET), peritumoral edematous, and necrotic and non‐ET on MRI; with performance quantified via a 5‐fold cross‐validated Dice coefficient. MRI scans were evaluated through the open‐source quality control tool MRQy, to yield 13 IQMs per scan. The Pearson correlation coefficient was computed between whole tumor (WT) dice values and IQM measures in the training cohorts to identify quality measures most correlated with segmentation performance. Each selected IQM was used to group MRI scans as “better” quality (BQ) or “worse” quality (WQ), via relative thresholding. Segmentation performance was re‐evaluated for the DenseNet model when (i) training on BQ MRI images with validation on WQ images, as well as (ii) training on WQ images, and validation on BQ images. Trends were further validated on independent test sets derived from the BraTS 2021 training cohorts.


For this study, multimodal MRI scans from the BraTS 2020 training cohorts were used to train the segmentation model and validated on independent test sets derived from the BraTS 2021 cohort. Among the selected IQMs, models trained on BQ images based on inhomogeneity measurements (coefficient of variance, coefficient of joint variation, coefficient of variation of the foreground patch) and the models trained on WQ images based on noise measurement peak signal‐to‐noise ratio (SNR) yielded significantly improved tumor segmentation accuracy compared to their inverse models.


Our results suggest that a significant correlation may exist between specific MR IQMs and DenseNet‐based brain tumor segmentation performance. The selection of MRI scans for model training based on IQMs may yield more accurate and generalizable models in unseen validation.

more » « less
Author(s) / Creator(s):
 ;  ;  ;  ;  ;  ;  ;  
Publisher / Repository:
Wiley Blackwell (John Wiley & Sons)
Date Published:
Journal Name:
Medical Physics
Medium: X Size: p. 4898-4906
p. 4898-4906
Sponsoring Org:
National Science Foundation
More Like this
  1. Abraham, Ajith ; Pllana, Sabri ; Casalino, Gabriella ; Ma, Kun ; Bajaj, Anu (Ed.)
    Precision in segmenting cardiac MR images is critical for accurately diagnosing cardiovascular diseases. Several deep learning models have been shown useful in segmenting the structure of the heart, such as atrium, ventricle and myocardium, in cardiac MR images. Given the diverse image quality in cardiac MRI scans from various clinical settings, it is currently uncertain how different levels of noise affect the precision of deep learning image segmentation. This uncertainty could potentially lead to bias in subsequent diagnoses. The goal of this study is to examine the effects of noise in cardiac MRI segmentation using deep learning. We employed the Automated Cardiac Diagnosis Challenge MRI dataset and augmented it with varying degrees of Rician noise during model training to test the model’s capability in segmenting heart structures. Three models, including TransUnet, SwinUnet, and Unet, were compared by calculating the SNR-Dice relations to evaluate the models' noise resilience. Results show that the TransUnet model, which combines CNN and Transformer architectures, demonstrated superior noise resilience. Noise augmentation during model training improved the models' noise resilience for segmentation. The findings under-score the critical role of deep learning models in adequately handling diverse noise conditions for the segmentation of heart structures in cardiac images. 
    more » « less
  2. Purpose

    An image filter designed for reconstructing cerebrovascular trees from MR images is described. Current imaging techniques capture major cerebral vessels reliably, but often fail to detect small vessels, whose contrast is suppressed due to limited resolution, slow blood flow rate, and distortions around bifurcations or nonvascular structures. An incomplete view of angioarchitecture limits the information available to physicians.


    A novel Hessian‐based filter for contrast‐enhancement in MR angiography and venography for blood vessel reconstruction without introducing dangling segments is presented. We quantify filter performance with receiver‐operating‐characteristic and dice‐similarity‐coefficient analysis. Total extracted vascular length, number‐of‐segments, volume, surface‐to‐distance, and positional error are calculated for validation.


    Reconstruction of cerebrovascular trees from MR images of six volunteers show that the new filter renders more complete representations of subject‐specific cerebrovascular networks. Validation with phantom models shows the filter correctly detects blood vessels across all length scales without failing at bifurcations or distorting diameters.


    The novel filter can potentially improve the diagnosis of cerebrovascular diseases by delivering metrics and anatomy of the vasculature. It also facilitates the automated analysis of large datasets by computing biometrics free of operator subjectivity. The high quality reconstruction enables computational mesh generation for subject‐specific hemodynamic simulations. Magn Reson Med 77:398–410, 2017. © 2016 Wiley Periodicals, Inc.

    more » « less
  3. 7T magnetic resonance imaging (MRI) has the potential to drive our understanding of human brain function through new contrast and enhanced resolution. Whole brain segmentation is a key neuroimaging technique that allows for region-by-region analysis of the brain. Segmentation is also an important preliminary step that provides spatial and volumetric information for running other neuroimaging pipelines. Spatially localized atlas network tiles (SLANT) is a popular 3D convolutional neural network (CNN) tool that breaks the whole brain segmentation task into localized sub-tasks. Each sub-task involves a specific spatial location handled by an independent 3D convolutional network to provide high resolution whole brain segmentation results. SLANT has been widely used to generate whole brain segmentations from structural scans acquired on 3T MRI. However, the use of SLANT for whole brain segmentation from structural 7T MRI scans has not been successful due to the inhomogeneous image contrast usually seen across the brain in 7T MRI. For instance, we demonstrate the mean percent difference of SLANT label volumes between a 3T scan-rescan is approximately 1.73%, whereas its 3T-7T scan-rescan counterpart has higher differences around 15.13%. Our approach to address this problem is to register the whole brain segmentation performed on 3T MRI to 7T MRI and use this information to finetune SLANT for structural 7T MRI. With the finetuned SLANT pipeline, we observe a lower mean relative difference in the label volumes of ~8.43% acquired from structural 7T MRI data. Dice similarity coefficient between SLANT segmentation on the 3T MRI scan and the after finetuning SLANT segmentation on the 7T MRI increased from 0.79 to 0.83 with p<0.01. These results suggest finetuning of SLANT is a viable solution for improving whole brain segmentation on high resolution 7T structural imaging. 
    more » « less
  4. While gliomas have become the most common cancerous brain tumors, manual diagnoses from 3D MRIs are time-consuming and possibly inconsistent when conducted by different radiotherapists, which leads to the pressing demand for automatic segmentation of brain tumors. State-of-the-art approaches employ FCNs to automatically segment the MRI scans. In particular, 3D U-Net has achieved notable performance and motivated a series of subsequent works. However, their significant size and heavy computation have impeded their actual deployment. Although there exists a body of literature on the compression of CNNs using low-precision representations, they either focus on storage reduction without computational improvement or cause severe performance degradation. In this article, we propose a CNN training algorithm that approximates weights and activations using non-negative integers along with trained affine mapping functions. Moreover, our approach allows the dot-product operations to be performed in an integer-arithmetic manner and defers the floating-point decoding and encoding phases until the end of layers. Experimental results on BraTS 2018 show that our trained affine mapping approach achieves near full-precision dice accuracy with 8-bit weights and activations. In addition, we achieve a dice accuracy within 0.005 and 0.01 of the full-precision counterparts when using 4-bit and 2-bit precisions, respectively. 
    more » « less
  5. This paper presents a novel predictive model, MetaMorph, for metamorphic registration of images with appearance changes (i.e., caused by brain tumors). In contrast to previous learning-based registration methods that have little or no control over appearance-changes, our model introduces a new regularization that can effectively suppress the negative effects of appearance changing areas. In particular, we develop a piecewise regularization on the tangent space of diffeomorphic transformations (also known as initial velocity fields) via learned segmentation maps of abnormal regions. The geometric transformation and appearance changes are treated as joint tasks that are mutually beneficial. Our model MetaMorph is more robust and accurate when searching for an optimal registration solution under the guidance of segmentation, which in turn improves the segmentation performance by providing appropriately augmented training labels. We validate MetaMorph on real 3D human brain tumor magnetic resonance imaging (MRI) scans. Experimental results show that our model outperforms the state-of-the-art learning-based registration models. The proposed MetaMorph has great potential in various image-guided clinical interventions, e.g., real-time image-guided navigation systems for tumor removal surgery. 
    more » « less