scholarly journals Sharp loss: a new loss function for radiotherapy dose prediction based on fully convolutional networks

2021 ◽  
Vol 20 (1) ◽  
Author(s):  
Xue Bai ◽  
Jie Zhang ◽  
Binbing Wang ◽  
Shengye Wang ◽  
Yida Xiang ◽  
...  
Keyword(s):  
Loss Function ◽  
Target Volume ◽  
Left Breast ◽  
Absolute Difference ◽  
Rank Test ◽  
Convolutional Networks ◽  
Dose Prediction ◽  
Fully Convolutional Networks ◽  
The Mean ◽  
Radiotherapy Dose

Abstract Background Neural-network methods have been widely used for the prediction of dose distributions in radiotherapy. However, the prediction accuracy of existing methods may be degraded by the problem of dose imbalance. In this work, a new loss function is proposed to alleviate the dose imbalance and achieve more accurate prediction results. The U-Net architecture was employed to build a prediction model. Our study involved a total of 110 patients with left-breast cancer, who were previously treated by volumetric-modulated arc radiotherapy. The patient dataset was divided into training and test subsets of 100 and 10 cases, respectively. We proposed a novel ‘sharp loss’ function, and a parameter γ was used to adjust the loss properties. The mean square error (MSE) loss and the sharp loss with different γ values were tested and compared using the Wilcoxon signed-rank test. Results The sharp loss achieved superior dose prediction results compared to those of the MSE loss. The best performance with the MSE loss and the sharp loss was obtained when the parameter γ was set to 100. Specifically, the mean absolute difference values for the planning target volume were 318.87 ± 30.23 for the MSE loss versus 144.15 ± 16.27 for the sharp loss with γ = 100 (p < 0.05). The corresponding values for the ipsilateral lung, the heart, the contralateral lung, and the spinal cord were 278.99 ± 51.68 versus 198.75 ± 61.38 (p < 0.05), 216.99 ± 44.13 versus 144.86 ± 43.98 (p < 0.05), 125.96 ± 66.76 versus 111.86 ± 47.19 (p > 0.05), and 194.30 ± 14.51 versus 168.58 ± 25.97 (p < 0.05), respectively. Conclusions The sharp loss function could significantly improve the accuracy of radiotherapy dose prediction.

Comparing of two dimensional and three dimensional fully convolutional networks for radiotherapy dose prediction in left-sided breast cancer

2021 ◽  
Vol 104 (3) ◽  
pp. 003685042110381
Author(s):  
Xue Bai ◽  
Ze Liu ◽  
Jie Zhang ◽  
Shengye Wang ◽  
Qing Hou ◽  
...  
Keyword(s):  
Three Dimensional ◽  
Region Of Interest ◽  
Dice Similarity Coefficient ◽  
Two Dimensional ◽  
Dimensional Model ◽  
Dose Distributions ◽  
Convolutional Networks ◽  
Dose Prediction ◽  
Fully Convolutional Networks ◽  
Percentage Difference

Fully convolutional networks were developed for predicting optimal dose distributions for patients with left-sided breast cancer and compared the prediction accuracy between two-dimensional and three-dimensional networks. Sixty cases treated with volumetric modulated arc radiotherapy were analyzed. Among them, 50 cases were randomly chosen to conform the training set, and the remaining 10 were to construct the test set. Two U-Net fully convolutional networks predicted the dose distributions, with two-dimensional and three-dimensional convolution kernels, respectively. Computed tomography images, delineated regions of interest, or their combination were considered as input data. The accuracy of predicted results was evaluated against the clinical dose. Most types of input data retrieved a similar dose to the ground truth for organs at risk ( p > 0.05). Overall, the two-dimensional model had higher performance than the three-dimensional model ( p < 0.05). Moreover, the two-dimensional region of interest input provided the best prediction results regarding the planning target volume mean percentage difference (2.40 ± 0.18%), heart mean percentage difference (4.28 ± 2.02%), and the gamma index at 80% of the prescription dose are with tolerances of 3 mm and 3% (0.85 ± 0.03), whereas the two-dimensional combined input provided the best prediction regarding ipsilateral lung mean percentage difference (4.16 ± 1.48%), lung mean percentage difference (2.41 ± 0.95%), spinal cord mean percentage difference (0.67 ± 0.40%), and 80% Dice similarity coefficient (0.94 ± 0.01). Statistically, the two-dimensional combined inputs achieved higher prediction accuracy regarding 80% Dice similarity coefficient than the two-dimensional region of interest input (0.94 ± 0.01 vs 0.92 ± 0.01, p < 0.05). The two-dimensional data model retrieves higher performance than its three-dimensional counterpart for dose prediction, especially when using region of interest and combined inputs.

scholarly journals Dual Path Attention Net for Remote Sensing Semantic Image Segmentation

2020 ◽  
Vol 9 (10) ◽  
pp. 571
Author(s):  
Jinglun Li ◽  
Jiapeng Xiu ◽  
Zhengqiu Yang ◽  
Chen Liu
Keyword(s):  
Remote Sensing ◽  
Spatial Information ◽  
Semantic Segmentation ◽  
Remote Sensing Images ◽  
Convolutional Networks ◽  
Fully Convolutional Networks ◽  
Augmentation Strategies ◽  
Rich Information ◽  
The Mean ◽  
The Rich

Semantic segmentation plays an important role in being able to understand the content of remote sensing images. In recent years, deep learning methods based on Fully Convolutional Networks (FCNs) have proved to be effective for the sematic segmentation of remote sensing images. However, the rich information and complex content makes the training of networks for segmentation challenging, and the datasets are necessarily constrained. In this paper, we propose a Convolutional Neural Network (CNN) model called Dual Path Attention Network (DPA-Net) that has a simple modular structure and can be added to any segmentation model to enhance its ability to learn features. Two types of attention module are appended to the segmentation model, one focusing on spatial information the other focusing upon the channel. Then, the outputs of these two attention modules are fused to further improve the network’s ability to extract features, thus contributing to more precise segmentation results. Finally, data pre-processing and augmentation strategies are used to compensate for the small number of datasets and uneven distribution. The proposed network was tested on the Gaofen Image Dataset (GID). The results show that the network outperformed U-Net, PSP-Net, and DeepLab V3+ in terms of the mean IoU by 0.84%, 2.54%, and 1.32%, respectively.

scholarly journals Independent verification of treatment planning system calculations

Nukleonika ◽  
2021 ◽  
Vol 66 (2) ◽  
pp. 47-53
Author(s):  
Edyta Dąbrowska-Szewczyk ◽  
Anna Zawadzka ◽  
Beata Brzozowska ◽  
Agnieszka Walewska ◽  
Paweł Kukołowicz
Keyword(s):  
Treatment Planning ◽  
Three Dimensional ◽  
Conformal Radiotherapy ◽  
Target Volume ◽  
Planning System ◽  
Treatment Planning System ◽  
Rank Test ◽  
Average Dose ◽  
Independent Verification ◽  
The Mean

Abstract Purpose According to the available international recommendations, at least one independent verification of the calculations of number of monitor unit (MU) is required for every patient treated by teleradiotherapy. The aim of this study was to estimate the differences of dose distributions calculated with two treatment planning systems: Eclipse (Varian) and Oncentra MasterPlan (Elekta). Materials and methods The analysis was performed for 280 three-dimensional conformal radiotherapy treatment (3D-CRT) plans with photon beams from Varian accelerators: CL 600C/D X6 MV (109 plans), CL 2300C/D X6 MV (43 plans), and CL 2300C/D X15 MV (128 plans). The mean doses in the planning target volume (PTV) and doses at the isocenter point obtained with Eclipse and Oncentra MasterPlan (OMP) were compared with Wilcoxon matched-pairs signed rank test. Additionally, the treatment planning system (TPS) calculations were compared with dosimetric measurements performed in the inhomogeneous phantom. Results Data were analysed for 6 MV plans and for 15 MV plans separately, independently of the treatment machine. The dose values calculated in Eclipse were significantly (p <0.001) higher compared to calculations of OMP system. The average difference of the mean dose to PTV was (1.4 ± 1.0)% for X6 MV and (2.5 ± 0.6)% for X15 MV. Average dose disparities at the isocenter point were (1.3 ± 1.9)% and (2.1 ± 1.0)% for X6 MV and X15 MV beams, respectively. The largest differences were observed in lungs, air cavities, and bone structures. Moreover the variation in dosimetric measurements was less as compared to Eclipse calculations. Conclusions OMP calculations were introduced as the independent MU verification tool with the first action level range equal to 3.5%.

scholarly journals Weakly-Supervised Image Semantic Segmentation Based on Superpixel Region Merging

2019 ◽  
Vol 3 (2) ◽  
pp. 31 ◽  
Author(s):  
Quanchun Jiang ◽  
Olamide Timothy Tawose ◽  
Songwen Pei ◽  
Xiaodong Chen ◽  
Linhua Jiang ◽  
...  
Keyword(s):  
Neural Network ◽  
Image Annotation ◽  
Semantic Segmentation ◽  
Region Merging ◽  
Segmentation Method ◽  
Convolutional Networks ◽  
Fully Convolutional Networks ◽  
The Mean ◽  
Weakly Supervised ◽  
The Relationship

In this paper, we propose a semantic segmentation method based on superpixel region merging and convolutional neural network (CNN), referred to as regional merging neural network (RMNN). Image annotation has always been an important role in weakly-supervised semantic segmentation. Most methods use manual labeling. In this paper, super-pixels with similar features are combined using the relationship between each pixel after super-pixel segmentation to form a plurality of super-pixel blocks. Rough predictions are generated by the fully convolutional networks (FCN) so that certain super-pixel blocks will be labeled. We perceive and find other positive areas in an iterative way through the marked areas. This reduces the feature extraction vector and reduces the data dimension due to super-pixels. The algorithm not only uses superpixel merging to narrow down the target’s range but also compensates for the lack of weakly-supervised semantic segmentation at the pixel level. In the training of the network, we use the method of region merging to improve the accuracy of contour recognition. Our extensive experiments demonstrated the effectiveness of the proposed method with the PASCAL VOC 2012 dataset. In particular, evaluation results show that the mean intersection over union (mIoU) score of our method reaches as high as 44.6%. Because the cavity convolution is in the pooled downsampling operation, it does not degrade the network’s receptive field, thereby ensuring the accuracy of image semantic segmentation. The findings of this work thus open the door to leveraging the dilated convolution to improve the recognition accuracy of small objects.

scholarly journals Deep RetinaNet for Dynamic Left Ventricle Detection in Multiview Echocardiography Classification

2020 ◽  
Vol 2020 ◽  
pp. 1-6 ◽  
Author(s):  
Meijun Yang ◽  
Xiaoyan Xiao ◽  
Zhi Liu ◽  
Longkun Sun ◽  
Wei Guo ◽  
...  
Keyword(s):  
Left Ventricle ◽  
Three Dimensional ◽  
Detection Algorithm ◽  
Convolutional Networks ◽  
Fully Convolutional Networks ◽  
Dimensional Reconstruction ◽  
Feature Pyramid ◽  
The Mean ◽  
Accuracy Rates

Background. Currently, echocardiography has become an essential technology for the diagnosis of cardiovascular diseases. Accurate classification of apical two-chamber (A2C), apical three-chamber (A3C), and apical four-chamber (A4C) views and the precise detection of the left ventricle can significantly reduce the workload of clinicians and improve the reproducibility of left ventricle segmentation. In addition, left ventricle detection is significant for the three-dimensional reconstruction of the heart chambers. Method. RetinaNet is a one-stage object detection algorithm that can achieve high accuracy and efficiency at the same time. RetinaNet is mainly composed of the residual network (ResNet), the feature pyramid network (FPN), and two fully convolutional networks (FCNs); one FCN is for the classification task, and the other is for the border regression task. Results. In this paper, we use the classification subnetwork to classify A2C, A3C, and A4C images and use the regression subnetworks to detect the left ventricle simultaneously. We display not only the position of the left ventricle on the test image but also the view category on the image, which will facilitate the diagnosis. We used the mean intersection-over-union (mIOU) as an index to measure the performance of left ventricle detection and the accuracy as an index to measure the effect of the classification of the three different views. Our study shows that both classification and detection effects are noteworthy. The classification accuracy rates of A2C, A3C, and A4C are 1.000, 0.935, and 0.989, respectively. The mIOU values of A2C, A3C, and A4C are 0.858, 0.794, and 0.838, respectively.

scholarly journals Convolutional neural networks for automated CMR image segmentation in rats with myocardial infarcts

2020 ◽  
Author(s):  
Andrea Gondova ◽  
Magdalena Zurek ◽  
Johan Karlsson ◽  
Leif Hultin ◽  
Tobias Noeske ◽  
...  
Keyword(s):  
Magnetic Resonance Images ◽  
Project Teams ◽  
Absolute Difference ◽  
Cardiovascular Research ◽  
Mean Absolute Difference ◽  
Systolic Volume ◽  
Convolutional Networks ◽  
Fully Convolutional Networks ◽  
End Diastolic Volume ◽  
End Systolic Volume

AbstractIn translational cardiovascular research, delineation of left ventricle (LV) in magnetic resonance images is a crucial step in assessing heart’s function. Performed manually, this task is time-consuming and prone to inter- and intra-reader variability. Here we report first AI-based tool for segmentation of rat cardiovascular MRI. The method is an ensemble of fully convolutional networks and can quantify clinically relevant measures: end-diastolic volume (EDV), end-systolic volume (ESV), and ejection fraction (EF) automatically.Overall, our method reaches Dice score of 0.93 on the independent test set. The mean absolute difference of segmented volumes between automated and manual segmentation is 22.5μL for EDV, 13.6μL for ESV, and for EF 2.9%. Our work demonstrates the value of AI in development of tools that will significantly reduce time spent on repetitive work and result in increased efficiency of reporting data to project teams.

scholarly journals THE LAND COVER CLASSIFICATION USING A FEATURE PYRAMID NETWORKS ARCHITECTURE FROM SATELLITE IMAGERY

2020 ◽  
Vol XLIII-B3-2020 ◽  
pp. 241-246
Author(s):  
Q. Zhang ◽  
Y. Zhang ◽  
P. Yang ◽  
Y. Meng ◽  
S. Zhuo ◽  
...  
Keyword(s):  
Land Cover ◽  
Transfer Learning ◽  
Satellite Imagery ◽  
Loss Function ◽  
Land Cover Classification ◽  
Deep Convolutional Neural Networks ◽  
Convolutional Networks ◽  
Fully Convolutional Networks ◽  
Sensing Applications ◽  
Feature Pyramid

Abstract. Extracting land cover information from satellite imagery is of great importance for the task of automated monitoring in various remote sensing applications. Deep convolutional neural networks make this task more feasible, but they are limited by the small dataset of annotated images. In this paper, we present a fully convolutional networks architecture, FPN-VGG, that combines Feature Pyramid Networks and VGG. In order to accomplish the task of land cover classification, we create a land cover dataset of pixel-wise annotated images, and employ a transfer learning step and the variant dice loss function to promote the performance of FPN-VGG. The results indicate that FPN-VGG shows more competence for land cover classification comparing with other state-of-the-art fully convolutional networks. The transfer learning and dice loss function are beneficial to improve the performance of on the small and unbalanced dataset. Our best model on the dataset gets an overall accuracy of 82.9%, an average F1 score of 66.0% and an average IoU of 52.7%.

scholarly journals Small and Dim Ship Target Detection Based on Sea-Sky-Line

2019 ◽  
Vol 37 (1) ◽  
pp. 35-40 ◽  
Author(s):  
Yaohui Hu ◽  
Ke Zhang ◽  
Chao Xing
Keyword(s):  
Target Detection ◽  
Detection Method ◽  
Detection Algorithm ◽  
Sea Surface ◽  
Line Detection ◽  
Convolutional Networks ◽  
Fully Convolutional Networks ◽  
Noise Point ◽  
The Mean ◽  
Potential Area

In order to solve the problem of small and dim ship target detection under complex sea-sky background, we propose a target detection algorithm based on sea-sky line detection. Firstly, the paper locates the sea-sky-line based on fully convolutional networks, through which target potential area can be determined and disturbance can be excluded. Then the method based on the mean of four detection gradient is adopted to detect the small and dim ship target. The simulation results show that the method of sea-sky-line detection based on fully convolutional networks can overcome the disadvantages of the traditional methods and is suitable for complex background. The detection method proposed can filter the white noise point on the sea surface and thus can reduce false alarm, through which the detection of small and dim ship can be completed well.

Dosimetric comparison of fractionated radiosurgery plans using frameless Gamma Knife ICON and CyberKnife systems with linear accelerator–based radiosurgery plans for multiple large brain metastases

2020 ◽  
Vol 132 (5) ◽  
pp. 1473-1479 ◽  
Author(s):  
Eun Young Han ◽  
He Wang ◽  
Dershan Luo ◽  
Jing Li ◽  
Xin Wang
Keyword(s):  
Brain Metastases ◽  
Gamma Knife ◽  
Linear Accelerator ◽  
Treatment Time ◽  
Target Volume ◽  
Planning System ◽  
Treatment Planning System ◽  
Normal Brain ◽  
Large Brain ◽  
The Mean

OBJECTIVEFor patients with multiple large brain metastases with at least 1 target volume larger than 10 cm3, multifractionated stereotactic radiosurgery (MF-SRS) has commonly been delivered with a linear accelerator (LINAC). Recent advances of Gamma Knife (GK) units with kilovolt cone-beam CT and CyberKnife (CK) units with multileaf collimators also make them attractive choices. The purpose of this study was to compare the dosimetry of MF-SRS plans deliverable on GK, CK, and LINAC and to discuss related clinical issues.METHODSTen patients with 2 or more large brain metastases who had been treated with MF-SRS on LINAC were identified. The median planning target volume was 18.31 cm3 (mean 21.31 cm3, range 3.42–49.97 cm3), and the median prescribed dose was 27.0 Gy (mean 26.7 Gy, range 21–30 Gy), administered in 3 to 5 fractions. Clinical LINAC treatment plans were generated using inverse planning with intensity modulation on a Pinnacle treatment planning system (version 9.10) for the Varian TrueBeam STx system. GK and CK planning were retrospectively performed using Leksell GammaPlan version 10.1 and Accuray Precision version 1.1.0.0 for the CK M6 system. Tumor coverage, Paddick conformity index (CI), gradient index (GI), and normal brain tissue receiving 4, 12, and 20 Gy were used to compare plan quality. Net beam-on time and approximate planning time were also collected for all cases.RESULTSPlans from all 3 modalities satisfied clinical requirements in target coverage and normal tissue sparing. The mean CI was comparable (0.79, 0.78, and 0.76) for the GK, CK, and LINAC plans. The mean GI was 3.1 for both the GK and the CK plans, whereas the mean GI of the LINAC plans was 4.1. The lower GI of the GK and CK plans would have resulted in significantly lower normal brain volumes receiving a medium or high dose. On average, GK and CK plans spared the normal brain volume receiving at least 12 Gy and 20 Gy by approximately 20% in comparison with the LINAC plans. However, the mean beam-on time of GK (∼ 64 minutes assuming a dose rate of 2.5 Gy/minute) plans was significantly longer than that of CK (∼ 31 minutes) or LINAC (∼ 4 minutes) plans.CONCLUSIONSAll 3 modalities are capable of treating multiple large brain lesions with MF-SRS. GK has the most flexible workflow and excellent dosimetry, but could be limited by the treatment time. CK has dosimetry comparable to that of GK with a consistent treatment time of approximately 30 minutes. LINAC has a much shorter treatment time, but residual rotational error could be a concern.

Sign in / Sign up

Export Citation Format

Share Document