Tree-Encoded Conditional Random Fields for Image Synthesis

On the proper use of structural similarity for the robust evaluation of medical image synthesis models

PURPOSE

To propose good practices for using the structural similarity metric (SSIM) and reporting its value. SSIM is one of the most popular image quality metrics in use in the medical image synthesis community because of its alleged superiority over voxel-by-voxel measurements like the average error or the peak signal noise ratio (PSNR). It has seen massive adoption since its introduction, but its limitations are often overlooked. Notably, SSIM is designed to work on a strictly positive intensity scale, which is generally not the case in medical imaging. Common intensity scales such as the Houndsfield units (HU) contain negative numbers, and they can also be introduced by image normalization techniques such as the z-normalization.

METHODS

We created a series of experiments to quantify the impact of negative values in the SSIM computation. Specifically, we trained a 3D U-Net to synthesize T2 weighted MRI from T1 weighted MRI using the BRATS 2018 dataset. SSIM was computed on the synthetic images with a shifted dynamic range. Next, to evaluate the suitability of SSIM as a loss function on images with negative values, it was used as a loss function to synthesize z-normalized images. Finally, the difference between 2D SSIM and 3D SSIM was investigated using multiple 2D U-Nets trained on different planes of the images.

RESULTS

The impact of the misuse of the SSIM was quantified; it was established that it introduces a large downward bias in the computed SSIM. It also introduces a small random error that can change the relative ranking of models. The exact values for this bias and error depend on the quality and the intensity histogram of the synthetic images. Although small, the reported error is significant considering the small SSIM difference between state-of-the-art models. It was shown therefore that SSIM cannot be used as a loss function when images contain negative values due to major errors in the gradient calculation, resulting in under-performing models. 2D SSIM was also found to be overestimated in 2D image synthesis models when computed along the plane of synthesis, due to the discontinuities between slices that is typical of 2D synthesis methods.

CONCLUSION

Various types of misuse of the SSIM were identified and their impact was quantified. Based on the findings, this paper proposes good practices when using SSIM, such as reporting the average over the volume of the image containing tissue and appropriately defining the dynamic range. This article is protected by copyright. All rights reserved.

mustGAN: multi-stream Generative Adversarial Networks for MR Image Synthesis

Multi-contrast MRI protocols increase the level of morphological information available for diagnosis. Yet, the number and quality of contrasts are limited in practice by various factors including scan time and patient motion. Synthesis of missing or corrupted contrasts from other high-quality ones can alleviate this limitation. When a single target contrast is of interest, common approaches for multi-contrast MRI involve either one-to-one or many-to-one synthesis methods depending on their input. One-to-one methods take as input a single source contrast, and they learn a latent representation sensitive to unique features of the source. Meanwhile, many-to-one methods receive multiple distinct sources, and they learn a shared latent representation more sensitive to common features across sources. For enhanced image synthesis, we propose a multi-stream approach that aggregates information across multiple source images via a mixture of multiple one-to-one streams and a joint many-to-one stream. The complementary feature maps generated in the one-to-one streams and the shared feature maps generated in the many-to-one stream are combined with a fusion block. The location of the fusion block is adaptively modified to maximize task-specific performance. Quantitative and radiological assessments on T₁,- T₂-, PD-weighted, and FLAIR images clearly demonstrate the superior performance of the proposed method compared to previous state-of-the-art one-to-one and many-to-one methods.

Missing MRI Pulse Sequence Synthesis Using Multi-Modal Generative Adversarial Network

Magnetic resonance imaging (MRI) is being increasingly utilized to assess, diagnose, and plan treatment for a variety of diseases. The ability to visualize tissue in varied contrasts in the form of MR pulse sequences in a single scan provides valuable insights to physicians, as well as enabling automated systems performing downstream analysis. However, many issues like prohibitive scan time, image corruption, different acquisition protocols, or allergies to certain contrast materials may hinder the process of acquiring multiple sequences for a patient. This poses challenges to both physicians and automated systems since complementary information provided by the missing sequences is lost. In this paper, we propose a variant of generative adversarial network (GAN) capable of leveraging redundant information contained within multiple available sequences in order to generate one or more missing sequences for a patient scan. The proposed network is designed as a multi-input, multi-output network which combines information from all the available pulse sequences and synthesizes the missing ones in a single forward pass. We demonstrate and validate our method on two brain MRI datasets each with four sequences, and show the applicability of the proposed method in simultaneously synthesizing all missing sequences in any possible scenario where either one, two, or three of the four sequences may be missing. We compare our approach with competing unimodal and multi-modal methods, and show that we outperform both quantitatively and qualitatively.

Brain lesion segmentation through image synthesis and outlier detection

Cerebral small vessel disease (SVD) can manifest in a number of ways. Many of these result in hyperintense regions visible on T2-weighted magnetic resonance (MR) images. The automatic segmentation of these lesions has been the focus of many studies. However, previous methods tended to be limited to certain types of pathology, as a consequence of either restricting the search to the white matter, or by training on an individual pathology. Here we present an unsupervised abnormality detection method which is able to detect abnormally hyperintense regions on FLAIR regardless of the underlying pathology or location. The method uses a combination of image synthesis, Gaussian mixture models and one class support vector machines, and needs only be trained on healthy tissue. We evaluate our method by comparing segmentation results from 127 subjects with SVD with three established methods and report significantly superior performance across a number of metrics.

Random forest regression for magnetic resonance image synthesis

By choosing different pulse sequences and their parameters, magnetic resonance imaging (MRI) can generate a large variety of tissue contrasts. This very flexibility, however, can yield inconsistencies with MRI acquisitions across datasets or scanning sessions that can in turn cause inconsistent automated image analysis. Although image synthesis of MR images has been shown to be helpful in addressing this problem, an inability to synthesize both T2-weighted brain images that include the skull and FLuid Attenuated Inversion Recovery (FLAIR) images has been reported. The method described herein, called REPLICA, addresses these limitations. REPLICA is a supervised random forest image synthesis approach that learns a nonlinear regression to predict intensities of alternate tissue contrasts given specific input tissue contrasts. Experimental results include direct image comparisons between synthetic and real images, results from image analysis tasks on both synthetic and real images, and comparison against other state-of-the-art image synthesis methods. REPLICA is computationally fast, and is shown to be comparable to other methods on tasks they are able to perform. Additionally REPLICA has the capability to synthesize both T2-weighted images of the full head and FLAIR images, and perform intensity standardization between different imaging datasets.