Search

Deep learning (DL) models for medical image segmentation are highly influenced by intensity variations of input images and lack generalization due to primarily utilizing pixels' intensity information for inference. Acquiring sufficient training data is another challenge limiting models' applications. We proposed to leverage the consistency of organs' anatomical shape and position information in medical images. We introduced a framework leveraging recurring anatomical patterns through global binary masks for organ segmentation. Two scenarios were studied.1) Global binary masks were the only model's (i.e. U-Net) input, forcing exclusively encoding organs' position and shape information for segmentation/localization.2) Global binary masks were incorporated as an additional channel functioning as position/shape clues to mitigate training data scarcity. Two datasets of the brain and heart CT images with their ground-truth were split into (26:10:10) and (12:3:5) for training, validation, and test respectively. Training exclusively on global binary masks led to Dice scores of 0.77(0.06) and 0.85(0.04), with the average Euclidian distance of 3.12(1.43)mm and 2.5(0.93)mm relative to the center of mass of the ground truth for the brain and heart structures respectively. The outcomes indicate that a surprising degree of position and shape information is encoded through global binary masks. Incorporating global binary masks led to significantly higher accuracy relative to the model trained on only CT images in small subsets of training data; the performance improved by 4.3-125.3% and 1.3-48.1% for 1-8 training cases of the brain and heart datasets respectively. The findings imply the advantages of utilizing global binary masks for building generalizable models and to compensate for training data scarcity.

Adaptive radiotherapy (ART), especially online ART, effectively accounts for positioning errors and anatomical changes. One key component of online ART is accurately and efficiently delineating organs at risk (OARs) and targets on online images, such as CBCT, to meet the online demands of plan evaluation and adaptation. Deep learning (DL)-based automatic segmentation has gained great success in segmenting planning CT, but its applications to CBCT yielded inferior results due to the low image quality and limited available contour labels for training. To overcome these obstacles to online CBCT segmentation, we propose a registration-guided DL (RgDL) segmentation framework that integrates image registration algorithms and DL segmentation models. The registration algorithm generates initial contours, which were used as guidance by DL model to obtain accurate final segmentations. We had two implementations the proposed framework--Rig-RgDL (Rig for rigid body) and Def-RgDL (Def for deformable)--with rigid body (RB) registration or deformable image registration (DIR) as the registration algorithm respectively and U-Net as DL model architecture. The two implementations of RgDL framework were trained and evaluated on seven OARs in an institutional clinical Head and Neck (HN) dataset. Compared to the baseline approaches using the registration or the DL alone, RgDL achieved more accurate segmentation, as measured by higher mean Dice similarity coefficients (DSC) and other distance-based metrics. Rig-RgDL achieved a DSC of 84.5% on seven OARs on average, higher than RB or DL alone by 4.5% and 4.7%. The DSC of Def-RgDL is 86.5%, higher than DIR or DL alone by 2.4% and 6.7%. The inference time took by the DL model to generate final segmentations of seven OARs is less than one second in RgDL. The resulting segmentation accuracy and efficiency show the promise of applying RgDL framework for online ART., Comment: 16 pages, 6 figures

Purpose: To develop a model to generate volumetric dose distribution from two isodose surfaces (iso-surfaces), and to interactively tune dose distribution by iso-surface dragging. Methods: We model volumetric dose distribution as analytical extension of two iso-surfaces with the extension variables as distances to iso-surfaces. We built a 3D lookup table (LUT) which are generated based on clinical dose distributions. Two LUT tables store the mean and standard deviation of voxel dose values of clinical doses and binned as distance to 100% iso-surface, reference iso-surface and reference dose level. The process of interactive tuning starts from a given base plan. A user drags iso-surface for a desired carving. Our method responds with tuned dose. The derivation of tuned dose follows two steps. Dose is extended from the two user-desired iso-surfaces (eg.100% and 50%) to the whole patient volume by table lookup, using distances to two iso-surfaces and reference dose level as keys. Then we fine tune the extended dose by a correction strategy utilizing the information of base plan. Results: We validated this method on coplanar VMAT doses of post-operative prostate plans. The LUT was populated by dose distributions of 27 clinical plans. We optimized two plans with different rectum sparing for an independent case to mimic the process of dose tuning. The plan with less rectum sparing is set as base plan. The 50% iso-surface of the more-sparing plan is defined as the desired iso-surface input. The dose output by our method (expansion and correction) agrees with the more-sparing plan obtained by optimization, in terms of gamma (97.2%), DVH and profiles. The overall dose reconstruction time is within two seconds. Conclusion: We developed a distance-to-isosurface based volumetric dose reconstruction method, and applied it to interactive tuning with iso-surface dragging., Comment: 19 pages, 10 figures

Deep learning-based fluence map prediction(DL-FMP) method has been reported in the literature, which generated fluence maps for desired dose by deep neural network(DNN)-based inverse mapping. We hypothesized that DL-FMP is similar to general fluence map optimization(FMO) because it's theoretically based on a general inverse mapping. We designed four experiments to validate the generalizability of DL-FMP to other types of plans apart from the training data, which contained only clinical head and neck(HN) full-arc VMAT plans. The first three experiments quantified the generalizability of DL-FMP to multiple anatomical sites, different delivery modalities, and various degree of modulation(DOM), respectively. The fourth experiment explored the generalizability and stability to infeasible dose inputs. Results of the first experiment manifested that DL-FMP can generalize to lung, liver, esophagus and prostate, with gamma passing rates (GPR) higher than 95%(2%/2mm). The second experiment showed that DL-FMP can generalize to partial-arc plans and predict partial-arc fluences. GPR(3mm/3%) ranged from 96% to 99%. DL-FMP cannot generate fluence maps in discrete beam angles for IMRT input. But the predicted dose still agreed with ground truth dose with 93% GPR(5%/5mm). The third experiment demonstrated that DL-FMP can generalize to various DOMs, with GPRs(3%/3mm) ranged in 94%-98%. Moreover, the DOM of predicted fluence maps correlated to the optimality of the input dose accordingly. The fourth experiment exemplified that DL-FMP can make stable predictions for infeasible dose input. In conclusion, we validated that DL-FMP can generalize to plans for multiple anatomical sites, plans of different delivery modalities and plans with various DOM. It can also make stable prediction for infeasible input.

We propose using deep reinforcement learning to solve dynamic stochastic general equilibrium models. Agents are represented by deep artificial neural networks and learn to solve their dynamic optimisation problem by interacting with the model environment, of which they have no a priori knowledge. Deep reinforcement learning offers a flexible yet principled way to model bounded rationality within this general class of models. We apply our proposed approach to a classical model from the adaptive learning literature in macroeconomics which looks at the interaction of monetary and fiscal policy. We find that, contrary to adaptive learning, the artificially intelligent household can solve the model in all policy regimes.

Partly due to the use of exhaustive-annotated data, deep networks have achieved impressive performance on medical image segmentation. Medical imaging data paired with noisy annotation are, however, ubiquitous, but little is known about the effect of noisy annotation on deep learning-based medical image segmentation. We studied the effects of noisy annotation in the context of mandible segmentation from CT images. First, 202 images of Head and Neck cancer patients were collected from our clinical database, where the organs-at-risk were annotated by one of 12 planning dosimetrists. The mandibles were roughly annotated as the planning avoiding structure. Then, mandible labels were checked and corrected by a physician to get clean annotations. At last, by varying the ratios of noisy labels in the training data, deep learning-based segmentation models were trained, one for each ratio. In general, a deep network trained with noisy labels had worse segmentation results than that trained with clean labels, and fewer noisy labels led to better segmentation. When using 20% or less noisy cases for training, no significant difference was found on the prediction performance between the models trained by noisy or clean. This study suggests that deep learning-based medical image segmentation is robust to noisy annotations to some extent. It also highlights the importance of labeling quality in deep learning

We propose a generalization of the linear panel quantile regression model to accommodate both \textit{sparse} and \textit{dense} parts: sparse means while the number of covariates available is large, potentially only a much smaller number of them have a nonzero impact on each conditional quantile of the response variable; while the dense part is represent by a low-rank matrix that can be approximated by latent factors and their loadings. Such a structure poses problems for traditional sparse estimators, such as the $\ell_1$-penalised Quantile Regression, and for traditional latent factor estimator, such as PCA. We propose a new estimation procedure, based on the ADMM algorithm, consists of combining the quantile loss function with $\ell_1$ \textit{and} nuclear norm regularization. We show, under general conditions, that our estimator can consistently estimate both the nonzero coefficients of the covariates and the latent low-rank matrix. Our proposed model has a "Characteristics + Latent Factors" Asset Pricing Model interpretation: we apply our model and estimator with a large-dimensional panel of financial data and find that (i) characteristics have sparser predictive power once latent factors were controlled (ii) the factors and coefficients at upper and lower quantiles are different from the median., Comment: forthcoming at the Annals of Statistics

Breast ultrasound (US) is an effective imaging modality for breast cancer detec-tion and diagnosis. The structural characteristics of breast lesion play an im-portant role in Computer-Aided Diagnosis (CAD). In this paper, a novel struc-ture-aware triplet path networks (SATPN) was designed to integrate classifica-tion and two image reconstruction tasks to achieve accurate diagnosis on US im-ages with small training dataset. Specifically, we enhance clinically-approved breast lesion structure characteristics though converting original breast US imag-es to BIRADS-oriented feature maps (BFMs) with a distance-transformation coupled Gaussian filter. Then, the converted BFMs were used as the inputs of SATPN, which performed lesion classification task and two unsupervised stacked convolutional Auto-Encoder (SCAE) networks for benign and malignant image reconstruction tasks, independently. We trained the SATPN with an alter-native learning strategy by balancing image reconstruction error and classification label prediction error. At the test stage, the lesion label was determined by the weighted voting with reconstruction error and label prediction error. We com-pared the performance of the SATPN with TPN using original image as input and our previous developed semi-supervised deep learning methods using BFMs as inputs. Experimental results on two breast US datasets showed that SATPN ranked the best among the three networks, with classification accuracy around 93.5%. These findings indicated that SATPN is promising for effective breast US lesion CAD using small datasets., Comment: arXiv admin note: substantial text overlap with arXiv:1904.01076

Data in the form of networks are increasingly available in a variety of areas, yet statistical models allowing for parameter estimates with desirable statistical properties for sparse networks remain scarce. To address this, we propose the Sparse $\beta$-Model (S$\beta$M), a new network model that interpolates the celebrated Erd\H{o}s-R\'enyi model and the $\beta$-model that assigns one different parameter to each node. By a novel reparameterization of the $\beta$-model to distinguish global and local parameters, our S$\beta$M can drastically reduce the dimensionality of the $\beta$-model by requiring some of the local parameters to be zero. We derive the asymptotic distribution of the maximum likelihood estimator of the S$\beta$M when the support of the parameter vector is known. When the support is unknown, we formulate a penalized likelihood approach with the $\ell_0$-penalty. Remarkably, we show via a monotonicity lemma that the seemingly combinatorial computational problem due to the $\ell_0$-penalty can be overcome by assigning nonzero parameters to those nodes with the largest degrees. We further show that a $\beta$-min condition guarantees our method to identify the true model and provide excess risk bounds for the estimated parameters. The estimation procedure enjoys good finite sample properties as shown by simulation studies. The usefulness of the S$\beta$M is further illustrated via the analysis of a microfinance take-up example., Comment: 36 pages

Breast ultrasound (US) is an effective imaging modality for breast cancer detection and diagnosis. US computer-aided diagnosis (CAD) systems have been developed for decades and have employed either conventional hand-crafted features or modern automatic deep-learned features, the former relying on clinical experience and the latter demanding large datasets. In this paper, we have developed a novel BIRADS-SDL network that integrates clinically-approved breast lesion characteristics (BIRADS features) into semi-supervised deep learning (SDL) to achieve accurate diagnoses with a small training dataset. Breast US images are converted to BIRADS-oriented feature maps (BFMs) using a distance-transformation coupled with a Gaussian filter. Then, the converted BFMs are used as the input of an SDL network, which performs unsupervised stacked convolutional auto-encoder (SCAE) image reconstruction guided by lesion classification. We trained the BIRADS-SDL network with an alternative learning strategy by balancing reconstruction error and classification label prediction error. We compared the performance of the BIRADS-SDL network with conventional SCAE and SDL methods that use the original images as inputs, as well as with an SCAE that use BFMs as inputs. Experimental results on two breast US datasets show that BIRADS-SDL ranked the best among the four networks, with classification accuracy around 92.00% and 83.90% on two datasets. These findings indicate that BIRADS-SDL could be promising for effective breast US lesion CAD using small datasets.

Data cleaning consumes about 80% of the time spent on data analysis for clinical research projects. This is a much bigger problem in the era of big data and machine learning in the field of medicine where large volumes of data are being generated. We report an initial effort towards automated patient data cleaning using deep learning: the standardization of organ labeling in radiation therapy. Organs are often labeled inconsistently at different institutions (sometimes even within the same institution) and at different time periods, which poses a problem for clinical research, especially for multi-institutional collaborative clinical research where the acquired patient data is not being used effectively. We developed a convolutional neural network (CNN) to automatically identify each organ in the CT image and then label it with the standardized nomenclature presented at AAPM Task Group 263. We tested this model on the CT images of 54 patients with prostate and 100 patients with head and neck cancer who previously received radiation therapy. The model achieved 100% accuracy in detecting organs and assigning standardized labels for the patients tested. This work shows the feasibility of using deep learning in patient data cleaning that enables standardized datasets to be generated for effective intra- and interinstitutional collaborative clinical research., Comment: 17 pages, 7 figures, 3 tables, 39 references