Learning to segment fetal brain tissue from noisy annotations

Automatic fetal brain tissue segmentation can enhance the quantitative assessment of brain development at this critical stage. Deep learning methods represent the state of the art in medical image segmentation and have also achieved impressive results in brain segmentation. However, effective training of a deep learning model to perform this task requires a large number of training images to represent the rapid development of the transient fetal brain structures. On the other hand, manual multi-label segmentation of a large number of 3D images is prohibitive. To address this challenge, we segmented 272 training images, covering 19-39 gestational weeks, using an automatic multi-atlas segmentation strategy based on deformable registration and probabilistic atlas fusion, and manually corrected large errors in those segmentations. Since this process generated a large training dataset with noisy segmentations, we developed a novel label smoothing procedure and a loss function to train a deep learning model with smoothed noisy segmentations. Our proposed methods properly account for the uncertainty in tissue boundaries. We evaluated our method on 23 manually-segmented test images of a separate set of fetuses. Results show that our method achieves an average Dice similarity coefficient of 0.893 and 0.916 for the transient structures of younger and older fetuses, respectively. Our method generated results that were significantly more accurate than several state-of-the-art methods including nnU-Net that achieved the closest results to our method. Our trained model can serve as a valuable tool to enhance the accuracy and reproducibility of fetal brain analysis in MRI.

Diffusion Tensor Estimation with Transformer Neural Networks

Diffusion tensor imaging (DTI) is the most widely used tool for studying brain white matter development and degeneration. However, standard DTI estimation methods depend on a large number of high-quality measurements. This would require long scan times and can be particularly difficult to achieve with certain patient populations such as neonates. Here, we propose a method that can accurately estimate the diffusion tensor from only six diffusion-weighted measurements. Our method achieves this by learning to exploit the relationships between the diffusion signals and tensors in neighboring voxels. Our model is based on transformer networks, which represent the state of the art in modeling the relationship between signals in a sequence. In particular, our model consists of two such networks. The first network estimates the diffusion tensor based on the diffusion signals in a neighborhood of voxels. The second network provides more accurate tensor estimations by learning the relationships between the diffusion signals as well as the tensors estimated by the first network in neighboring voxels. Our experiments with three datasets show that our proposed method achieves highly accurate estimations of the diffusion tensor and is significantly superior to three competing methods. Estimations produced by our method with six measurements are comparable with those of standard estimation methods with 30-88 measurements. Hence, our method promises shorter scan times and more reliable assessment of brain white matter, particularly in non-cooperative patients such as neonates and infants.

Calibrated Diffusion Tensor Estimation

It is highly desirable to know how uncertain a model's predictions are, especially for models that are complex and hard to understand as in deep learning. Although there has been a growing interest in using deep learning methods in diffusion-weighted MRI, prior works have not addressed the issue of model uncertainty. Here, we propose a deep learning method to estimate the diffusion tensor and compute the estimation uncertainty. Data-dependent uncertainty is computed directly by the network and learned via loss attenuation. Model uncertainty is computed using Monte Carlo dropout. We also propose a new method for evaluating the quality of predicted uncertainties. We compare the new method with the standard least-squares tensor estimation and bootstrap-based uncertainty computation techniques. Our experiments show that when the number of measurements is small the deep learning method is more accurate and its uncertainty predictions are better calibrated than the standard methods. We show that the estimation uncertainties computed by the new method can highlight the model's biases, detect domain shift, and reflect the strength of noise in the measurements. Our study shows the importance and practical value of modeling prediction uncertainties in deep learning-based diffusion MRI analysis.

Convolution-Free Medical Image Segmentation using Transformers

Like other applications in computer vision, medical image segmentation has been most successfully addressed using deep learning models that rely on the convolution operation as their main building block. Convolutions enjoy important properties such as sparse interactions, weight sharing, and translation equivariance. These properties give convolutional neural networks (CNNs) a strong and useful inductive bias for vision tasks. In this work we show that a different method, based entirely on self-attention between neighboring image patches and without any convolution operations, can achieve competitive or better results. Given a 3D image block, our network divides it into $n^3$ 3D patches, where $n=3 \text{ or } 5$ and computes a 1D embedding for each patch. The network predicts the segmentation map for the center patch of the block based on the self-attention between these patch embeddings. We show that the proposed model can achieve segmentation accuracies that are better than the state of the art CNNs on three datasets. We also propose methods for pre-training this model on large corpora of unlabeled images. Our experiments show that with pre-training the advantage of our proposed network over CNNs can be significant when labeled training data is small.

A machine learning-based method for estimating the number and orientations of major fascicles in diffusion-weighted magnetic resonance imaging

Multi-compartment modeling of diffusion-weighted magnetic resonance imaging measurements is necessary for accurate brain connectivity analysis. Existing methods for estimating the number and orientations of fascicles in an imaging voxel either depend on non-convex optimization techniques that are sensitive to initialization and measurement noise, or are prone to predicting spurious fascicles. In this paper, we propose a machine learning-based technique that can accurately estimate the number and orientations of fascicles in a voxel. Our method can be trained with either simulated or real diffusion-weighted imaging data. Our method estimates the angle to the closest fascicle for each direction in a set of discrete directions uniformly spread on the unit sphere. This information is then processed to extract the number and orientations of fascicles in a voxel. On realistic simulated phantom data with known ground truth, our method predicts the number and orientations of crossing fascicles more accurately than several existing methods. It also leads to more accurate tractography. On real data, our method is better than or compares favorably with standard methods in terms of robustness to measurement down-sampling and also in terms of expert quality assessment of tractography results.

Critical Assessment of Transfer Learning for Medical Image Segmentation with Fully Convolutional Neural Networks

Transfer learning is widely used for training machine learning models. Here, we study the role of transfer learning for training fully convolutional networks (FCNs) for medical image segmentation. Our experiments show that although transfer learning reduces the training time on the target task, the improvement in segmentation accuracy is highly task/data-dependent. Larger improvements in accuracy are observed when the segmentation task is more challenging and the target training data is smaller. We observe that convolutional filters of an FCN change little during training for medical image segmentation, and still look random at convergence. We further show that quite accurate FCNs can be built by freezing the encoder section of the network at random values and only training the decoder section. At least for medical image segmentation, this finding challenges the common belief that the encoder section needs to learn data/task-specific representations. We examine the evolution of FCN representations to gain a better insight into the effects of transfer learning on the training dynamics. Our analysis shows that although FCNs trained via transfer learning learn different representations than FCNs trained with random initialization, the variability among FCNs trained via transfer learning can be as high as that among FCNs trained with random initialization. Moreover, feature reuse is not restricted to the early encoder layers; rather, it can be more significant in deeper layers. These findings offer new insights and suggest alternative ways of training FCNs for medical image segmentation.

A Deep Attentive Convolutional Neural Network for Automatic Cortical Plate Segmentation in Fetal MRI

Fetal cortical plate segmentation is essential in quantitative analysis of fetal brain maturation and cortical folding. Manual segmentation of the cortical plate, or manual refinement of automatic segmentations is tedious and time consuming, and automatic segmentation of the cortical plate is challenged by the relatively low resolution of the reconstructed fetal brain MRI scans compared to the thin structure of the cortical plate, partial voluming, and the wide range of variations in the morphology of the cortical plate as the brain matures during gestation. To reduce the burden of manual refinement of segmentations, we have developed a new and powerful deep learning segmentation method that exploits new deep attentive modules with mixed kernel convolutions within a fully convolutional neural network architecture that utilizes deep supervision and residual connections. Quantitative evaluation based on several performance measures and expert evaluations show that our method outperformed several state-of-the-art deep models for segmentation, as well as a state-of-the-art multi-atlas segmentation technique. In particular, we achieved average Dice similarity coefficient of 0.87, average Hausdroff distance of 0.96mm, and average symmetric surface difference of 0.28mm in cortical plate segmentation on reconstructed fetal brain MRI scans of fetuses scanned in the gestational age range of 16 to 39 weeks. By generating accurate cortical plate segmentations in less than 2 minutes, our method can facilitate and accelerate large-scale studies on normal and altered fetal brain cortical maturation and folding.

Improving Calibration and Out-of-Distribution Detection in Medical Image Segmentation with Convolutional Neural Networks

Convolutional Neural Networks (CNNs) are powerful medical image segmentation models. In this study, we address some of the main unresolved issues regarding these models. Specifically, training of these models on small medical image datasets is still challenging, with many studies promoting techniques such as transfer learning. Moreover, these models are infamous for producing over-confident predictions and for failing silently when presented with out-of-distribution (OOD) data at test time. In this paper, we advocate for training on heterogeneous data, i.e., training a single model on several different datasets, spanning several different organs of interest and different imaging modalities. We show that not only a single CNN learns to automatically recognize the context and accurately segment the organ of interest in each context, but also that such a joint model often has more accurate and better-calibrated predictions than dedicated models trained separately on each dataset. We also show that training on heterogeneous data can outperform transfer learning. For detecting OOD data, we propose a method based on spectral analysis of CNN feature maps. We show that different datasets, representing different imaging modalities and/or different organs of interest, have distinct spectral signatures, which can be used to identify whether or not a test image is similar to the images used to train a model. We show that this approach is far more accurate than OOD detection based on prediction uncertainty. The methods proposed in this paper contribute significantly to improving the accuracy and reliability of CNN-based medical image segmentation models.

Deep learning with noisy labels: exploring techniques and remedies in medical image analysis

Supervised training of deep learning models requires large labeled datasets. There is a growing interest in obtaining such datasets for medical image analysis applications. However, the impact of label noise has not received sufficient attention. Recent studies have shown that label noise can significantly impact the performance of deep learning models in many machine learning and computer vision applications. This is especially concerning for medical applications, where datasets are typically small, labeling requires domain expertise and suffers from high inter- and intra-observer variability, and erroneous predictions may influence decisions that directly impact human health. In this paper, we first review the state-of-the-art in handling label noise in deep learning. Then, we review studies that have dealt with label noise in deep learning for medical image analysis. Our review shows that recent progress on handling label noise in deep learning has gone largely unnoticed by the medical image analysis community. To help achieve a better understanding of the extent of the problem and its potential remedies, we conducted experiments with three medical imaging datasets with different types of label noise. Based on the results of these experiments and our review of the literature, we make recommendations on methods that can be used to alleviate the effects of different types of label noise on deep models trained for medical image analysis. We hope that this article helps the medical image analysis researchers and developers in choosing and devising new techniques that effectively handle label noise in deep learning.