arXiv subjects

Jonas Teuwen

Publications and source records attributed to Jonas Teuwen.

54 records · Page 3Linked to original sources

arXiv2021

WeakSTIL: Weak whole-slide image level stromal tumor infiltrating lymphocyte scores are all you need

We present WeakSTIL, an interpretable two-stage weak label deep learning pipeline for scoring the percentage of stromal tumor infiltrating lymphocytes (sTIL%) in H&E-stained whole-slide images (WSIs) of breast cancer tissue. The sTIL% score is a prognostic and predictive biomarker for many solid tumor types. However, due to the high labeling efforts and high intra- and interobserver variability within and between expert annotators, this biomarker is currently not used in routine clinical decision making. WeakSTIL compresses tiles of a WSI using a feature extractor pre-trained with self-supervised learning on unlabeled histopathology data and learns to predict precise sTIL% scores for each tile in the tumor bed by using a multiple instance learning regressor that only requires a weak WSI-level label. By requiring only a weak label, we overcome the large annotation efforts required to train currently existing TIL detection methods. We show that WeakSTIL is at least as good as other TIL detection methods when predicting the WSI-level sTIL% score, reaching a coefficient of determination of $0.45\pm0.15$ when compared to scores generated by an expert pathologist, and an AUC of $0.89\pm0.05$ when treating it as the clinically interesting sTIL-high vs sTIL-low classification task. Additionally, we show that the intermediate tile-level predictions of WeakSTIL are highly interpretable, which suggests that WeakSTIL pays attention to latent features related to the number of TILs and the tissue type. In the future, WeakSTIL may be used to provide consistent and interpretable sTIL% predictions to stratify breast cancer patients into targeted therapy arms.

eess.IV↗

arXiv2021

Sparse-shot Learning with Exclusive Cross-Entropy for Extremely Many Localisations

Object localisation, in the context of regular images, often depicts objects like people or cars. In these images, there is typically a relatively small number of objects per class, which usually is manageable to annotate. However, outside the setting of regular images, we are often confronted with a different situation. In computational pathology, digitised tissue sections are extremely large images, whose dimensions quickly exceed 250'000x250'000 pixels, where relevant objects, such as tumour cells or lymphocytes can quickly number in the millions. Annotating them all is practically impossible and annotating sparsely a few, out of many more, is the only possibility. Unfortunately, learning from sparse annotations, or sparse-shot learning, clashes with standard supervised learning because what is not annotated is treated as a negative. However, assigning negative labels to what are true positives leads to confusion in the gradients and biased learning. To this end, we present exclusive cross-entropy, which slows down the biased learning by examining the second-order loss derivatives in order to drop the loss terms corresponding to likely biased terms. Experiments on nine datasets and two different localisation tasks, detection with YOLLO and segmentation with Unet, show that we obtain considerable improvements compared to cross-entropy or focal loss, while often reaching the best possible performance for the model with only 10-40% of annotations.

cs.LG↗

arXiv2021

Automatic Breast Lesion Detection in Ultrafast DCE-MRI Using Deep Learning

Purpose: We propose a deep learning-based computer-aided detection (CADe) method to detect breast lesions in ultrafast DCE-MRI sequences. This method uses both the three-dimensional spatial information and temporal information obtained from the early-phase of the dynamic acquisition. Methods: The proposed CADe method, based on a modified 3D RetinaNet model, operates on ultrafast T1 weighted sequences, which are preprocessed for motion compensation, temporal normalization, and are cropped before passing into the model. The model is optimized to enable the detection of relatively small breast lesions in a screening setting, focusing on detection of lesions that are harder to differentiate from confounding structures inside the breast. Results: The method was developed based on a dataset consisting of 489 ultrafast MRI studies obtained from 462 patients containing a total of 572 lesions (365 malignant, 207 benign) and achieved a detection rate, sensitivity, and detection rate of benign lesions of 0.90 (0.876-0.934), 0.95 (0.934-0.980), and 0.81 (0.751-0.871) at 4 false positives per normal breast with 10-fold cross-testing, respectively. Conclusions: The deep learning architecture used for the proposed CADe application can efficiently detect benign and malignant lesions on ultrafast DCE-MRI. Furthermore, utilizing the less visible hard-to detect-lesions in training improves the learning process and, subsequently, detection of malignant breast lesions.

eess.IV↗

arXiv2021

Results of the 2020 fastMRI Challenge for Machine Learning MR Image Reconstruction

Accelerating MRI scans is one of the principal outstanding problems in the MRI research community. Towards this goal, we hosted the second fastMRI competition targeted towards reconstructing MR images with subsampled k-space data. We provided participants with data from 7,299 clinical brain scans (de-identified via a HIPAA-compliant procedure by NYU Langone Health), holding back the fully-sampled data from 894 of these scans for challenge evaluation purposes. In contrast to the 2019 challenge, we focused our radiologist evaluations on pathological assessment in brain images. We also debuted a new Transfer track that required participants to submit models evaluated on MRI scanners from outside the training set. We received 19 submissions from eight different groups. Results showed one team scoring best in both SSIM scores and qualitative radiologist evaluations. We also performed analysis on alternative metrics to mitigate the effects of background noise and collected feedback from the participants to inform future challenges. Lastly, we identify common failure modes across the submissions, highlighting areas of need for future research in the MRI reconstruction community.

eess.IV↗

arXiv2021

Deep learning reconstruction of digital breast tomosynthesis images for accurate breast density and patient-specific radiation dose estimation

The two-dimensional nature of mammography makes estimation of the overall breast density challenging, and estimation of the true patient-specific radiation dose impossible. Digital breast tomosynthesis (DBT), a pseudo-3D technique, is now commonly used in breast cancer screening and diagnostics. Still, the severely limited 3rd dimension information in DBT has not been used, until now, to estimate the true breast density or the patient-specific dose. This study proposes a reconstruction algorithm for DBT based on deep learning specifically optimized for these tasks. The algorithm, which we name DBToR, is based on unrolling a proximal-dual optimization method. The proximal operators are replaced with convolutional neural networks and prior knowledge is included in the model. This extends previous work on a deep learning-based reconstruction model by providing both the primal and the dual blocks with breast thickness information, which is available in DBT. Training and testing of the model were performed using virtual patient phantoms from two different sources. Reconstruction performance, and accuracy in estimation of breast density and radiation dose, were estimated, showing high accuracy (density <+/-3%; dose <+/-20%) without bias, significantly improving on the current state-of-the-art. This work also lays the groundwork for developing a deep learning-based reconstruction algorithm for the task of image interpretation by radiologists.

physics.med-ph↗

arXiv2020

Deep Learning-based Initialization of Iterative Reconstruction for Breast Tomosynthesis

Reconstruction of digital breast tomosynthesis is a challenging problem due to the limited angle data available in such systems. Due to memory limitations, deep learning-based methods can help improve these reconstructions, but can not (yet) attain sufficiently high resolution. In addition to this practical issue, questions remain on the possibility of such models introducing 'ghost' information from the training data that is not compatible with the projection data. To take advantage of some of the benefits of deep learning-based reconstructions while avoiding these limitations, we propose to use the low resolution deep learning-based reconstruction as an initialization of a regular high resolution iterative method. The network was trained using digital phantoms, some based on a mathematical model and some derived from patient dedicated breast CT scans. The output of this network was then used as initialization for 10 000 iterations of MLTR for nine patient based phantoms that were not included in the training. The same nine cases were also reconstructed without any initialization for comparison. The reconstructions including initialization were found to reach a lower mean squared error than those without, and visual inspection found much improved retrieval of the breast outline and depiction of the skin, confirming that adding the deep learning-based initialization adds valuable information to the reconstruction.

physics.med-ph↗

arXiv2020

Kernel of CycleGAN as a Principle homogeneous space

Unpaired image-to-image translation has attracted significant interest due to the invention of CycleGAN, a method which utilizes a combination of adversarial and cycle consistency losses to avoid the need for paired data. It is known that the CycleGAN problem might admit multiple solutions, and our goal in this paper is to analyze the space of exact solutions and to give perturbation bounds for approximate solutions. We show theoretically that the exact solution space is invariant with respect to automorphisms of the underlying probability spaces, and, furthermore, that the group of automorphisms acts freely and transitively on the space of exact solutions. We examine the case of zero `pure' CycleGAN loss first in its generality, and, subsequently, expand our analysis to approximate solutions for `extended' CycleGAN loss where identity loss term is included. In order to demonstrate that these results are applicable, we show that under mild conditions nontrivial smooth automorphisms exist. Furthermore, we provide empirical evidence that neural networks can learn these automorphisms with unexpected and unwanted results. We conclude that finding optimal solutions to the CycleGAN loss does not necessarily lead to the envisioned result in image-to-image translation tasks and that underlying hidden symmetries can render the result utterly useless.

cs.LG↗

arXiv2019

i-RIM applied to the fastMRI challenge

We, team AImsterdam, summarize our submission to the fastMRI challenge (Zbontar et al., 2018). Our approach builds on recent advances in invertible learning to infer models as presented in Putzky and Welling (2019). Both, our single-coil and our multi-coil model share the same basic architecture.

eess.IV↗

arXiv2019

Learned SIRT for Cone Beam Computed Tomography Reconstruction

We introduce the learned simultaneous iterative reconstruction technique (SIRT) for tomographic reconstruction. The learned SIRT algorithm is a deep learning based reconstruction method combining model knowledge with a learned component. The algorithm is trained by mapping raw measured data to the reconstruction results over several iterations. The Learned SIRT algorithm is applied to a cone beam geometry on a circular orbit, a challenging problem for learned methods due to its 3D geometry and its inherent inability to completely capture the patient anatomy. A comparison of 2D reconstructions is shown, where the learned SIRT approach produces reconstructions with superior peak signal to noise ratio (PSNR) and structural similarity (SSIM), compared to FBP, SIRT and U-net post-processing and similar PSNR and SSIM compared to the learned primal dual algorithm. Similar results are shown for cone beam geometry reconstructions of a 3D Shepp Logan phantom, where we obtain between 9.9 and 28.1 dB improvement over FBP with a substantial improvement in SSIM. Finally we show that our algorithm scales to clinically relevant problems, and performs well when applied to measurements of a physical phantom.

eess.IV↗

arXiv2018

Improving Breast Cancer Detection using Symmetry Information with Deep Learning

Convolutional Neural Networks (CNN) have had a huge success in many areas of computer vision and medical image analysis. However, there is still an immense potential for performance improvement in mammogram breast cancer detection Computer-Aided Detection (CAD) systems by integrating all the information that the radiologist utilizes, such as symmetry and temporal data. In this work, we proposed a patch based multi-input CNN that learns symmetrical difference to detect breast masses. The network was trained on a large-scale dataset of 28294 mammogram images. The performance was compared to a baseline architecture without symmetry context using Area Under the ROC Curve (AUC) and Competition Performance Metric (CPM). At candidate level, AUC value of 0.933 with 95% confidence interval of [0.920, 0.954] was obtained when symmetry information is incorporated in comparison with baseline architecture which yielded AUC value of 0.929 with [0.919, 0.947] confidence interval. By incorporating symmetrical information, although there was no a significant candidate level performance again (p = 0.111), we have found a compelling result at exam level with CPM value of 0.733 (p = 0.001). We believe that including temporal data, and adding benign class to the dataset could improve the detection performance.

cs.CV↗

arXiv2018

Deep Learning Framework for Digital Breast Tomosynthesis Reconstruction

Digital breast tomosynthesis is rapidly replacing digital mammography as the basic x-ray technique for evaluation of the breasts. However, the sparse sampling and limited angular range gives rise to different artifacts, which manufacturers try to solve in several ways. In this study we propose an extension of the Learned Primal-Dual algorithm for digital breast tomosynthesis. The Learned Primal-Dual algorithm is a deep neural network consisting of several `reconstruction blocks', which take in raw sinogram data as the initial input, perform a forward and a backward pass by taking projections and back-projections, and use a convolutional neural network to produce an intermediate reconstruction result which is then improved further by the successive reconstruction block. We extend the architecture by providing breast thickness measurements as a mask to the neural network and allow it to learn how to use this thickness mask. We have trained the algorithm on digital phantoms and the corresponding noise-free/noisy projections, and then tested the algorithm on digital phantoms for varying level of noise. Reconstruction performance of the algorithms was compared visually, using MSE loss and Structural Similarity Index. Results indicate that the proposed algorithm outperforms the baseline iterative reconstruction algorithm in terms of reconstruction quality for both breast edges and internal structures and is robust to noise.

physics.med-ph↗

arXiv2018

Vendor-independent soft tissue lesion detection using weakly supervised and unsupervised adversarial domain adaptation

Computer-aided detection aims to improve breast cancer screening programs by helping radiologists to evaluate digital mammography (DM) exams. DM exams are generated by devices from different vendors, with diverse characteristics between and even within vendors. Physical properties of these devices and postprocessing of the images can greatly influence the resulting mammogram. This results in the fact that a deep learning model trained on data from one vendor cannot readily be applied to data from another vendor. This paper investigates the use of tailored transfer learning methods based on adversarial learning to tackle this problem. We consider a database of DM exams (mostly bilateral and two views) generated by Hologic and Siemens vendors. We analyze two transfer learning settings: 1) unsupervised transfer, where Hologic data with soft lesion annotation at pixel level and Siemens unlabelled data are used to annotate images in the latter data; 2) weak supervised transfer, where exam level labels for images from the Siemens mammograph are available. We propose tailored variants of recent state-of-the-art methods for transfer learning which take into account the class imbalance and incorporate knowledge provided by the annotations at exam level. Results of experiments indicate the beneficial effect of transfer learning in both transfer settings. Notably, at 0.02 false positives per image, we achieve a sensitivity of 0.37, compared to 0.30 of a baseline with no transfer. Results indicate that using exam level annotations gives an additional increase in sensitivity.

cs.CV↗

arXiv2018

Automated soft tissue lesion detection and segmentation in digital mammography using a u-net deep learning network

Computer-aided detection or decision support systems aim to improve breast cancer screening programs by helping radiologists to evaluate digital mammography (DM) exams. Commonly such methods proceed in two steps: selection of candidate regions for malignancy, and later classification as either malignant or not. In this study, we present a candidate detection method based on deep learning to automatically detect and additionally segment soft tissue lesions in DM. A database of DM exams (mostly bilateral and two views) was collected from our institutional archive. In total, 7196 DM exams (28294 DM images) acquired with systems from three different vendors (General Electric, Siemens, Hologic) were collected, of which 2883 contained malignant lesions verified with histopathology. Data was randomly split on an exam level into training (50\%), validation (10\%) and testing (40\%) of deep neural network with u-net architecture. The u-net classifies the image but also provides lesion segmentation. Free receiver operating characteristic (FROC) analysis was used to evaluate the model, on an image and on an exam level. On an image level, a maximum sensitivity of 0.94 at 7.93 false positives (FP) per image was achieved. Similarly, per exam a maximum sensitivity of 0.98 at 7.81 FP per image was achieved. In conclusion, the method could be used as a candidate selection model with high accuracy and with the additional information of lesion segmentation.

cs.CV↗

arXiv2018

Student Beats the Teacher: Deep Neural Networks for Lateral Ventricles Segmentation in Brain MR

Ventricular volume and its progression are known to be linked to several brain diseases such as dementia and schizophrenia. Therefore accurate measurement of ventricle volume is vital for longitudinal studies on these disorders, making automated ventricle segmentation algorithms desirable. In the past few years, deep neural networks have shown to outperform the classical models in many imaging domains. However, the success of deep networks is dependent on manually labeled data sets, which are expensive to acquire especially for higher dimensional data in the medical domain. In this work, we show that deep neural networks can be trained on much-cheaper-to-acquire pseudo-labels (e.g., generated by other automated less accurate methods) and still produce more accurate segmentations compared to the quality of the labels. To show this, we use noisy segmentation labels generated by a conventional region growing algorithm to train a deep network for lateral ventricle segmentation. Then on a large manually annotated test set, we show that the network significantly outperforms the conventional region growing algorithm which was used to produce the training labels for the network. Our experiments report a Dice Similarity Coefficient (DSC) of $0.874$ for the trained network compared to $0.754$ for the conventional region growing algorithm ($p < 0.001$).

cs.CV↗

arXiv2018

On Maximum Focused Electric Energy in Bounded Regions

A general method is presented for determining the maximum electric energy in a bounded region of optical fields with given time-averaged flux of electromagnetic energy. Time-harmonic fields are considered whose plane wave expansion consists of propagating plane waves only, i.e., evanescent waves are excluded. The bounded region can be quite general: it can consist of finitely many points, or be a curve, a curved surface or a bounded volume. The optimum optical field is eigenfield corresponding to the maximum eigenvalue of a compact linear integral operator which depends on the bounded region. It is explained how these optimum fields can be realized by focusing appropriate pupil fields. The special case that the region is a circular disc perpendicular to the direction of optical axis is investigated by numerical simulations.

physics.optics↗

arXiv2016

On the integral kernels of derivatives of the Ornstein-Uhlenbeck semigroup

This paper presents a closed-form expression for the integral kernels associated with the derivatives of the Ornstein-Uhlenbeck semigroup $e^{tL}$. Our approach is to expand the Mehler kernel into Hermite polynomials and applying the powers $L^N$ of the Ornstein-Uhlenbeck operator to it, where we exploit the fact that the Hermite polynomials are eigenfunctions for $L$. As an application we give an alternative proof of the kernel estimates by Portal [2014], making all relevant quantities explicit.

math.AP↗

arXiv2016

$L^p$-$L^q$ off-diagonal estimates for the Ornstein--Uhlenbeck semigroup: some positive and negative results

We investigate $L^p(γ)$-$L^q(γ)$ off-diagonal estimates for the Ornstein-Uhlenbeck semigroup $(e^{tL})_{t > 0}$. For sufficiently large $t$ (quantified in terms of $p$ and $q$) these estimates hold in an unrestricted sense, while for sufficiently small $t$ they fail when restricted to maximal admissible balls and sufficiently small annuli. Our counterexample uses Mehler kernel estimates.

math.FA↗

arXiv2014

A note on the Gaussian maximal functions

This note presents a proof that the non-tangential maximal function of the Ornstein-Uhlenbeck semigroup is bounded almost surely by the Gaussian Hardy-Littlewood maximal function. In particular this entails improvement on a result by Pineda and Urbina who proved a similar result for a `trunctated' version of the non-tangential maximal function. We actually obtain boundedness of the maximal function on non-tangential cones of arbitrary aperture.

math.AP↗