Tobias Ortmaier

Max-Heinrich Laves, Sontje Ihler, Jacob F. Fast et al.

The consideration of predictive uncertainty in medical imaging with deep learning is of utmost importance. We apply estimation of both aleatoric and epistemic uncertainty by variational Bayesian inference with Monte Carlo dropout to regression tasks and show that predictive uncertainty is systematically underestimated. We apply $ σ$ scaling with a single scalar value; a simple, yet effective calibration method for both types of uncertainty. The performance of our approach is evaluated on a variety of common medical regression data sets using different state-of-the-art convolutional network architectures. In our experiments, $ σ$ scaling is able to reliably recalibrate predictive uncertainty. It is easy to implement and maintains the accuracy. Well-calibrated uncertainty in regression allows robust rejection of unreliable predictions or detection of out-of-distribution samples. Our source code is available at https://github.com/mlaves/well-calibrated-regression-uncertainty

Max-Heinrich Laves, Malte Tölle, Tobias Ortmaier

Uncertainty quantification in inverse medical imaging tasks with deep learning has received little attention. However, deep models trained on large data sets tend to hallucinate and create artifacts in the reconstructed output that are not anatomically present. We use a randomly initialized convolutional network as parameterization of the reconstructed image and perform gradient descent to match the observation, which is known as deep image prior. In this case, the reconstruction does not suffer from hallucinations as no prior training is performed. We extend this to a Bayesian approach with Monte Carlo dropout to quantify both aleatoric and epistemic uncertainty. The presented method is evaluated on the task of denoising different medical imaging modalities. The experimental results show that our approach yields well-calibrated uncertainty. That is, the predictive uncertainty correlates with the predictive error. This allows for reliable uncertainty estimates and can tackle the problem of hallucinations and artifacts in inverse medical imaging tasks.

Sontje Ihler, Max-Heinrich Laves, Tobias Ortmaier

Fast motion feedback is crucial in computer-aided surgery (CAS) on moving tissue. Image-assistance in safety-critical vision applications requires a dense tracking of tissue motion. This can be done using optical flow (OF). Accurate motion predictions at high processing rates lead to higher patient safety. Current deep learning OF models show the common speed vs. accuracy trade-off. To achieve high accuracy at high processing rates, we propose patient-specific fine-tuning of a fast model. This minimizes the domain gap between training and application data, while reducing the target domain to the capability of the lower complex, fast model. We propose to obtain training sequences pre-operatively in the operation room. We handle missing ground truth, by employing teacher-student learning. Using flow estimations from teacher model FlowNet2 we specialize a fast student model FlowNet2S on the patient-specific domain. Evaluation is performed on sequences from the Hamlyn dataset. Our student model shows very good performance after fine-tuning. Tracking accuracy is comparable to the teacher model at a speed up of factor six. Fine-tuning can be performed within minutes, making it feasible for the operation room. Our method allows to use a real-time capable model that was previously not suited for this task. This method is laying the path for improved patient-specific motion estimation in CAS.

Max-Heinrich Laves, Sontje Ihler, Karl-Philipp Kortmann et al.

The model uncertainty obtained by variational Bayesian inference with Monte Carlo dropout is prone to miscalibration. In this paper, different logit scaling methods are extended to dropout variational inference to recalibrate model uncertainty. Expected uncertainty calibration error (UCE) is presented as a metric to measure miscalibration. The effectiveness of recalibration is evaluated on CIFAR-10/100 and SVHN for recent CNN architectures. Experimental results show that logit scaling considerably reduce miscalibration by means of UCE. Well-calibrated uncertainty enables reliable rejection of uncertain predictions and robust detection of out-of-distribution data.

Max-Heinrich Laves, Sontje Ihler, Karl-Philipp Kortmann et al.

Model uncertainty obtained by variational Bayesian inference with Monte Carlo dropout is prone to miscalibration. The uncertainty does not represent the model error well. In this paper, temperature scaling is extended to dropout variational inference to calibrate model uncertainty. Expected uncertainty calibration error (UCE) is presented as a metric to measure miscalibration of uncertainty. The effectiveness of this approach is evaluated on CIFAR-10/100 for recent CNN architectures. Experimental results show, that temperature scaling considerably reduces miscalibration by means of UCE and enables robust rejection of uncertain predictions. The proposed approach can easily be derived from frequentist temperature scaling and yields well-calibrated model uncertainty. It is simple to implement and does not affect the model accuracy.

Max-Heinrich Laves, Sontje Ihler, Tobias Ortmaier

We evaluate two different methods for the integration of prediction uncertainty into diagnostic image classifiers to increase patient safety in deep learning. In the first method, Monte Carlo sampling is applied with dropout at test time to get a posterior distribution of the class labels (Bayesian ResNet). The second method extends ResNet to a probabilistic approach by predicting the parameters of the posterior distribution and sampling the final result from it (Variational ResNet).The variance of the posterior is used as metric for uncertainty.Both methods are trained on a data set of optical coherence tomography scans showing four different retinal conditions. Our results shown that cases in which the classifier predicts incorrectly correlate with a higher uncertainty. Mean uncertainty of incorrectly diagnosed cases was between 4.6 and 8.1 times higher than mean uncertainty of correctly diagnosed cases. Modeling of the prediction uncertainty in computer-aided diagnosis with deep learning yields more reliable results and is anticipated to increase patient safety.

Max-Heinrich Laves, Sontje Ihler, Tobias Ortmaier

We present deformable unsupervised medical image registration using a randomly-initialized deep convolutional neural network (CNN) as regularization prior. Conventional registration methods predict a transformation by minimizing dissimilarities between an image pair. The minimization is usually regularized with manually engineered priors, which limits the potential of the registration. By learning transformation priors from a large dataset, CNNs have achieved great success in deformable registration. However, learned methods are restricted to domain-specific data and the required amounts of medical data are difficult to obtain. Our approach uses the idea of deep image priors to combine convolutional networks with conventional registration methods based on manually engineered priors. The proposed method is applied to brain MRI scans. We show that our approach registers image pairs with state-of-the-art accuracy by providing dense, pixel-wise correspondence maps. It does not rely on prior training and is therefore not limited to a specific image domain.

Max-Heinrich Laves, Sontje Ihler, Lüder Alexander Kahrs et al.

Noise in speckle-prone optical coherence tomography tends to obfuscate important details necessary for medical diagnosis. In this paper, a denoising approach that preserves disease characteristics on retinal optical coherence tomography images in ophthalmology is presented. By combining a deep convolutional autoencoder with a priorly trained ResNet image classifier as regularizer, the perceptibility of delicate details is encouraged and only information-less background noise is filtered out. With our approach, higher peak signal-to-noise ratios with $ \mathrm{PSNR} = 31.2\,\mathrm{dB} $ and higher classification accuracy of $\mathrm{ACC} = 85.0\,\%$ can be achieved for denoised images compared to state-of-the-art denoising with $ \mathrm{PSNR} = 29.4\,\mathrm{dB} $ or $\mathrm{ACC} = 70.3\,\%$, depending on the method. It is shown that regularized autoencoders are capable of denoising retinal OCT images without blurring details of diseases.

Max-Heinrich Laves, Sontje Ihler, Lüder A. Kahrs et al.

According to the World Health Organization, 285 million people worldwide live with visual impairment. The most commonly used imaging technique for diagnosis in ophthalmology is optical coherence tomography (OCT). However, analysis of retinal OCT requires trained ophthalmologists and time, making a comprehensive early diagnosis unlikely. A recent study established a diagnostic tool based on convolutional neural networks (CNN), which was trained on a large database of retinal OCT images. The performance of the tool in classifying retinal conditions was on par to that of trained medical experts. However, the training of these networks is based on an enormous amount of labeled data, which is expensive and difficult to obtain. Therefore, this paper describes a method based on variational autoencoder regularization that improves classification performance when using a limited amount of labeled data. This work uses a two-path CNN model combining a classification network with an autoencoder (AE) for regularization. The key idea behind this is to prevent overfitting when using a limited training dataset size with small number of patients. Results show superior classification performance compared to a pre-trained and fully fine-tuned baseline ResNet-34. Clustering of the latent space in relation to the disease class is distinct. Neural networks for disease classification on OCTs can benefit from regularization using variational autoencoders when trained with limited amount of patient data. Especially in the medical imaging domain, data annotated by experts is expensive to obtain.

Max-Heinrich Laves, Sarah Latus, Jan Bergmeier et al.

Purpose: The facial recess is a delicate structure that must be protected in minimally invasive cochlear implant surgery. Current research estimates the drill trajectory by using endoscopy of the unique mastoid patterns. However, missing depth information limits available features for a registration to preoperative CT data. Therefore, this paper evaluates OCT for enhanced imaging of drill holes in mastoid bone and compares OCT data to original endoscopic images. Methods: A catheter-based OCT probe is inserted into a drill trajectory of a mastoid phantom in a translation-rotation manner to acquire the inner surface state. The images are undistorted and stitched to create volumentric data of the drill hole. The mastoid cell pattern is segmented automatically and compared to ground truth. Results: The mastoid pattern segmented on images acquired with OCT show a similarity of J = 73.6 % to ground truth based on endoscopic images and measured with the Jaccard metric. Leveraged by additional depth information, automated segmentation tends to be more robust and fail-safe compared to endoscopic images. Conclusion: The feasibility of using a clinically approved OCT probe for imaging the drill hole in cochlear implantation is shown. The resulting volumentric images provide additional information on the shape of caveties in the bone structure, which will be useful for image-to-patient registration and to estimate the drill trajectory. This will be another step towards safe minimally invasive cochlear implantation.

Max-Heinrich Laves, Lüder A. Kahrs, Tobias Ortmaier

In microsurgery, lasers have emerged as precise tools for bone ablation. A challenge is automatic control of laser bone ablation with 4D optical coherence tomography (OCT). OCT as high resolution imaging modality provides volumetric images of tissue and foresees information of bone position and orientation (pose) as well as thickness. However, existing approaches for OCT based laser ablation control rely on external tracking systems or invasively ablated artificial landmarks for tracking the pose of the OCT probe relative to the tissue. This can be superseded by estimating the scene flow caused by relative movement between OCT-based laser ablation system and patient. Therefore, this paper deals with 2.5D scene flow estimation of volumetric OCT images for application in laser ablation. We present a semi-supervised convolutional neural network based tracking scheme for subsequent 3D OCT volumes and apply it to a realistic semi-synthetic data set of ex vivo human temporal bone specimen. The scene flow is estimated in a two-stage approach. In the first stage, 2D lateral scene flow is computed on census-transformed en-face arguments-of-maximum intensity projections. Subsequent to this, the projections are warped by predicted lateral flow and 1D depth flow is estimated. The neural network is trained semi-supervised by combining error to ground truth and the reconstruction error of warped images with assumptions of spatial flow smoothness. Quantitative evaluation reveals a mean endpoint error of $ (4.7\pm{}3.5) $ voxel or $ 27.5 \pm 20.5 μ\mathrm{m} $ for scene flow estimation caused by simulated relative movement between the OCT probe and bone. The scene flow estimation for 4D OCT enables its use for markerless tracking of mastoid bone structures for image guidance in general, and automated laser ablation control.

Max-Heinrich Laves, Jens Bicker, Lüder A. Kahrs et al.

Purpose Automated segmentation of anatomical structures in medical image analysis is a prerequisite for autonomous diagnosis as well as various computer and robot aided interventions. Recent methods based on deep convolutional neural networks (CNN) have outperformed former heuristic methods. However, those methods were primarily evaluated on rigid, real-world environments. In this study, existing segmentation methods were evaluated for their use on a new dataset of transoral endoscopic exploration. Methods Four machine learning based methods SegNet, UNet, ENet and ErfNet were trained with supervision on a novel 7-class dataset of the human larynx. The dataset contains 536 manually segmented images from two patients during laser incisions. The Intersection-over-Union (IoU) evaluation metric was used to measure the accuracy of each method. Data augmentation and network ensembling were employed to increase segmentation accuracy. Stochastic inference was used to show uncertainties of the individual models. Patient-to-patient transfer was investigated using patient-specific fine-tuning. Results In this study, a weighted average ensemble network of UNet and ErfNet was best suited for the segmentation of laryngeal soft tissue with a mean IoU of 84.7 %. The highest efficiency was achieved by ENet with a mean inference time of 9.22 ms per image. It is shown that 10 additional images from a new patient are sufficient for patient-specific fine-tuning. Conclusion CNN-based methods for semantic segmentation are applicable to endoscopic images of laryngeal soft tissue. The segmentation can be used for active constraints or to monitor morphological changes and autonomously detect pathologies. Further improvements could be achieved by using a larger dataset or training the models in a self-supervised manner on additional unlabeled data.