Abstract P319: Can Deep Learning Find the Ischemic Core on CT? Transfer Learning From Pre-Trained MRI-Based Networks

Stroke ◽  
2021 ◽  
Vol 52 (Suppl_1) ◽  
Author(s):  
Yannan Yu ◽  
Soren Christensen ◽  
Yuan Xie ◽  
Enhao Gong ◽  
Maarten G Lansberg ◽  
...  
Keyword(s):  
Deep Learning ◽  
Ground Truth ◽  
Learning Model ◽  
Fine Tuning ◽  
Learning Models ◽  
Starting Point ◽  
Stroke Lesion ◽  
Ischemic Core ◽  
Deep Learning Model

Objective: Ischemic core prediction from CT perfusion (CTP) remains inaccurate compared with gold standard diffusion-weighted imaging (DWI). We evaluated if a deep learning model to predict the DWI lesion from MR perfusion (MRP) could facilitate ischemic core prediction on CTP. Method: Using the multi-center CRISP cohort of acute ischemic stroke patient with CTP before thrombectomy, we included patients with major reperfusion (TICI score≥2b), adequate image quality, and follow-up MRI at 3-7 days. Perfusion parameters including Tmax, mean transient time, cerebral blood flow (CBF), and cerebral blood volume were reconstructed by RAPID software. Core lab experts outlined the stroke lesion on the follow-up MRI. A previously trained MRI model in a separate group of patients was used as a starting point, which used MRP parameters as input and RAPID ischemic core on DWI as ground truth. We fine-tuned this model, using CTP parameters as input, and follow-up MRI as ground truth. Another model was also trained from scratch with only CTP data. 5-fold cross validation was used. Performance of the models was compared with ischemic core (rCBF≤30%) from RAPID software to identify the presence of a large infarct (volume>70 or >100ml). Results: 94 patients in the CRISP trial met the inclusion criteria (mean age 67±15 years, 52% male, median baseline NIHSS 18, median 90-day mRS 2). Without fine-tuning, the MRI model had an agreement of 73% in infarct >70ml, and 69% in >100ml; the MRI model fine-tuned on CT improved the agreement to 77% and 73%; The CT model trained from scratch had agreements of 73% and 71%; All of the deep learning models outperformed the rCBF segmentation from RAPID, which had agreements of 51% and 64%. See Table and figure. Conclusions: It is feasible to apply MRP-based deep learning model to CT. Fine-tuning with CTP data further improves the predictions. All deep learning models predict the stroke lesion after major recanalization better than thresholding approaches based on rCBF.

Abstract WP79: The Value of Pre-Training for Deep Learning Acute Stroke Triaging Models

Stroke ◽  
2020 ◽  
Vol 51 (Suppl_1) ◽  
Author(s):  
Yannan Yu ◽  
Yuan Xie ◽  
Enhao Gong ◽  
Thoralf thamm ◽  
Jiahong Ouyang ◽  
...  
Keyword(s):  
Deep Learning ◽  
Ground Truth ◽  
Residue Function ◽  
Fine Tuning ◽  
Wilcoxon Test ◽  
Lesion Volume ◽  
Learning Models ◽  
Wide Range ◽  
Training Strategies ◽  
Ischemic Core

Objective: We investigated if deep learning models are able to define the penumbra and ischemic core by comparing models from two training strategies (with and without pre-training) and clinical thresholding criteria (MRI parameter time-to-peak of the residue function [Tmax] and apparent diffusion coefficient [ADC]). Methods: We selected patients from two multicenter stroke trials, with baseline perfusion-weighted imaging (PWI) and diffusion-weighted imaging (DWI) and 3-7 day T2-FLAIR. Based on reperfusion rate calculated from baseline and 24 hr PWI, patients were grouped into unknown (no 24 hr PWI scan), minimal (≤20%), partial (20%-80%), and major (≥80%) reperfusion. Attention-gated U-net structure was selected for training, with eight image channels from baseline PWI/DWI as inputs and the infarct lesion manually segmented on T2-FLAIR as ground truth. Two training strategies were used: (1) training two models separately in minimal and major reperfusion patients; (2) pre-training a model using patients with partial and unknown reperfusion, then fine-tuning two models using minimal and major reperfusion patients, respectively. Prediction was evaluated by Dice score coefficient (DSC), and lesion volume error at an optimal threshold. In minimal and major reperfusion patients, the deep learning models and Tmax and ADC thresholding were compared using paired sample Wilcoxon test. Results: 182 patients were included (85 males, age 65±16 yrs, baseline NIHSS 15 IQR 10-19), with a breakdown of minimal/major/partial/unknown reperfusion status of 32/65/43/42 patients, respectively. The pre-training approach performed the best among all approaches (Table 1, Figure 1). Conclusion: Deep learning models to predict penumbra and ischemic core are best trained using general pre-training on a wide range of stroke cases followed by fine-tuning on the extreme cases. This method outperforms conventional DWI-PWI mismatch inspired thresholding approaches.

scholarly journals Performance Comparison of the Deep Learning and the Human Endoscopist for Bleeding Peptic Ulcer Disease

2021 ◽  
Author(s):  
Hsu-Heng Yen ◽  
Ping-Yu Wu ◽  
Pei-Yuan Su ◽  
Chia-Wei Yang ◽  
Yang-Yuan Chen ◽  
...  
Keyword(s):  
Peptic Ulcer ◽  
Deep Learning ◽  
Sensitivity And Specificity ◽  
Endoscopic Therapy ◽  
Learning Model ◽  
Ulcer Bleeding ◽  
Learning Models ◽  
Peptic Ulcer Bleeding ◽  
Ulcer Disease ◽  
Deep Learning Model

Abstract Purpose Management of peptic ulcer bleeding is clinically challenging. Accurate characterization of the bleeding during endoscopy is key for endoscopic therapy. This study aimed to assess whether a deep learning model can aid in the classification of bleeding peptic ulcer disease. Methods Endoscopic still images of patients (n = 1694) with peptic ulcer bleeding for the last 5 years were retrieved and reviewed. Overall, 2289 images were collected for deep learning model training, and 449 images were validated for the performance test. Two expert endoscopists classified the images into different classes based on their appearance. Four deep learning models, including Mobile Net V2, VGG16, Inception V4, and ResNet50, were proposed and pre-trained by ImageNet with the established convolutional neural network algorithm. A comparison of the endoscopists and trained deep learning model was performed to evaluate the model’s performance on a dataset of 449 testing images. Results The results first presented the performance comparisons of four deep learning models. The Mobile Net V2 presented the optimal performance of the proposal models. The Mobile Net V2 was chosen for further comparing the performance with the diagnostic results obtained by one senior and one novice endoscopists. The sensitivity and specificity were acceptable for the prediction of “normal” lesions in both 3-class and 4-class classifications. For the 3-class category, the sensitivity and specificity were 94.83% and 92.36%, respectively. For the 4-class category, the sensitivity and specificity were 95.40% and 92.70%, respectively. The interobserver agreement of the testing dataset of the model was moderate to substantial with the senior endoscopist. The accuracy of the determination of endoscopic therapy required and high-risk endoscopic therapy of the deep learning model was higher than that of the novice endoscopist. Conclusions In this study, the deep learning model performed better than inexperienced endoscopists. Further improvement of the model may aid in clinical decision-making during clinical practice, especially for trainee endoscopist.

scholarly journals IMG-22. A DEEP LEARNING MODEL FOR AUTOMATIC POSTERIOR FOSSA PEDIATRIC BRAIN TUMOR SEGMENTATION: A MULTI-INSTITUTIONAL STUDY

2020 ◽  
Vol 22 (Supplement_3) ◽  
pp. iii359-iii359
Author(s):  
Lydia Tam ◽  
Edward Lee ◽  
Michelle Han ◽  
Jason Wright ◽  
Leo Chen ◽  
...  
Keyword(s):  
Deep Learning ◽  
Posterior Fossa ◽  
Ground Truth ◽  
Learning Model ◽  
Tumor Segmentation ◽  
Dice Similarity Coefficient ◽  
Study Cohort ◽  
Navigation Systems ◽  
Tumor Types ◽  
Deep Learning Model

Abstract BACKGROUND Brain tumors are the most common solid malignancies in childhood, many of which develop in the posterior fossa (PF). Manual tumor measurements are frequently required to optimize registration into surgical navigation systems or for surveillance of nonresectable tumors after therapy. With recent advances in artificial intelligence (AI), automated MRI-based tumor segmentation is now feasible without requiring manual measurements. Our goal was to create a deep learning model for automated PF tumor segmentation that can register into navigation systems and provide volume output. METHODS 720 pre-surgical MRI scans from five pediatric centers were divided into training, validation, and testing datasets. The study cohort comprised of four PF tumor types: medulloblastoma, diffuse midline glioma, ependymoma, and brainstem or cerebellar pilocytic astrocytoma. Manual segmentation of the tumors by an attending neuroradiologist served as “ground truth” labels for model training and evaluation. We used 2D Unet, an encoder-decoder convolutional neural network architecture, with a pre-trained ResNet50 encoder. We assessed ventricle segmentation accuracy on a held-out test set using Dice similarity coefficient (0–1) and compared ventricular volume calculation between manual and model-derived segmentations using linear regression. RESULTS Compared to the ground truth expert human segmentation, overall Dice score for model performance accuracy was 0.83 for automatic delineation of the 4 tumor types. CONCLUSIONS In this multi-institutional study, we present a deep learning algorithm that automatically delineates PF tumors and outputs volumetric information. Our results demonstrate applied AI that is clinically applicable, potentially augmenting radiologists, neuro-oncologists, and neurosurgeons for tumor evaluation, surveillance, and surgical planning.

Transfer Learning of The ResNet-18 and DenseNet-121 Model Used to Diagnose Intracranial Hemorrhage in CT Scanning

2021 ◽  
Vol 27 ◽  
Author(s):  
Qi Zhou ◽  
Wenjie Zhu ◽  
Fuchen Li ◽  
Mingqing Yuan ◽  
Linfeng Zheng ◽  
...  
Keyword(s):  
Deep Learning ◽  
Transfer Learning ◽  
Intracranial Hemorrhage ◽  
Intraventricular Hemorrhage ◽  
Learning Model ◽  
Normal Group ◽  
Subdural Hemorrhage ◽  
Model Assessment ◽  
Learning Models ◽  
Deep Learning Model

Objective: To verify the ability of the deep learning model in identifying five subtypes and normal images in noncontrast enhancement CT of intracranial hemorrhage. Method: A total of 351 patients (39 patients in the normal group, 312 patients in the intracranial hemorrhage group) performed with intracranial hemorrhage noncontrast enhanced CT were selected, with 2768 images in total (514 images for the normal group, 398 images for the epidural hemorrhage group, 501 images for the subdural hemorrhage group, 497 images for the intraventricular hemorrhage group, 415 images for the cerebral parenchymal hemorrhage group, and 443 images for the subarachnoid hemorrhage group). Based on the diagnostic reports of two radiologists with more than 10 years of experience, the ResNet-18 and DenseNet-121 deep learning models were selected. Transfer learning was used. 80% of the data was used for training models, 10% was used for validating model performance against overfitting, and the last 10% was used for the final evaluation of the model. Assessment indicators included accuracy, sensitivity, specificity, and AUC values. Results: The overall accuracy of ResNet-18 and DenseNet-121 models were 89.64% and 82.5%, respectively. The sensitivity and specificity of identifying five subtypes and normal images were above 0.80. The sensitivity of DenseNet-121 model to recognize intraventricular hemorrhage and cerebral parenchymal hemorrhage was lower than 0.80, 0.73, and 0.76 respectively. The AUC values of the two deep learning models were above 0.9. Conclusion: The deep learning model can accurately identify the five subtypes of intracranial hemorrhage and normal images, and it can be used as a new tool for clinical diagnosis in the future.

scholarly journals A Physics-Infused Deep Learning Model for the Prediction of Refractive Indices and Its Use for the Large-Scale Screening of Organic Compound Space

2019 ◽  
Author(s):  
Mojtaba Haghighatlari ◽  
Gaurav Vishwakarma ◽  
Mohammad Atif Faiz Afzal ◽  
Johannes Hachmann
Keyword(s):  
Machine Learning ◽  
Deep Learning ◽  
Large Scale ◽  
Organic Molecules ◽  
Learning Model ◽  
Training Data ◽  
Refractive Indices ◽  
Learning Models ◽  
Deep Learning Model ◽  
Machine Learning Models

<div><div><div><p>We present a multitask, physics-infused deep learning model to accurately and efficiently predict refractive indices (RIs) of organic molecules, and we apply it to a library of 1.5 million compounds. We show that it outperforms earlier machine learning models by a significant margin, and that incorporating known physics into data-derived models provides valuable guardrails. Using a transfer learning approach, we augment the model to reproduce results consistent with higher-level computational chemistry training data, but with a considerably reduced number of corresponding calculations. Prediction errors of machine learning models are typically smallest for commonly observed target property values, consistent with the distribution of the training data. However, since our goal is to identify candidates with unusually large RI values, we propose a strategy to boost the performance of our model in the remoter areas of the RI distribution: We bias the model with respect to the under-represented classes of molecules that have values in the high-RI regime. By adopting a metric popular in web search engines, we evaluate our effectiveness in ranking top candidates. We confirm that the models developed in this study can reliably predict the RIs of the top 1,000 compounds, and are thus able to capture their ranking. We believe that this is the first study to develop a data-derived model that ensures the reliability of RI predictions by model augmentation in the extrapolation region on such a large scale. These results underscore the tremendous potential of machine learning in facilitating molecular (hyper)screening approaches on a massive scale and in accelerating the discovery of new compounds and materials, such as organic molecules with high-RI for applications in opto-electronics.</p></div></div></div>

Ground-truth uncertainty-aware metrics for machine learning applications on seismic image interpretation: Application to faults and horizon extraction

2020 ◽  
Vol 39 (10) ◽  
pp. 734-741
Author(s):  
Sébastien Guillon ◽  
Frédéric Joncour ◽  
Pierre-Emmanuel Barrallon ◽  
Laurent Castanié
Keyword(s):  
Deep Learning ◽  
Image Interpretation ◽  
Ground Truth ◽  
Learning Model ◽  
Seismic Interpretation ◽  
Data Set ◽  
Seismic Image ◽  
Machine Learning Applications ◽  
Deep Learning Model

We propose new metrics to measure the performance of a deep learning model applied to seismic interpretation tasks such as fault and horizon extraction. Faults and horizons are thin geologic boundaries (1 pixel thick on the image) for which a small prediction error could lead to inappropriately large variations in common metrics (precision, recall, and intersection over union). Through two examples, we show how classical metrics could fail to indicate the true quality of fault or horizon extraction. Measuring the accuracy of reconstruction of thin objects or boundaries requires introducing a tolerance distance between ground truth and prediction images to manage the uncertainties inherent in their delineation. We therefore adapt our metrics by introducing a tolerance function and illustrate their ability to manage uncertainties in seismic interpretation. We compare classical and new metrics through different examples and demonstrate the robustness of our metrics. Finally, we show on a 3D West African data set how our metrics are used to tune an optimal deep learning model.

scholarly journals Predicting glaucoma prior to its onset using deep learning

2019 ◽  
Author(s):  
Anshul Thakur ◽  
Michael Goldbaum ◽  
Siamak Yousefi
Keyword(s):  
Deep Learning ◽  
Visual Field ◽  
Optic Neuropathy ◽  
Disease Onset ◽  
Visual Fields ◽  
Learning Model ◽  
Glaucomatous Optic Neuropathy ◽  
Learning Models ◽  
Fundus Photographs ◽  
Deep Learning Model

AbstractPurposeTo assess the accuracy of deep learning models to predict glaucoma development from fundus photographs several years prior to disease onset.DesignA deep learning model for prediction of glaucomatous optic neuropathy or visual field abnormality from color fundus photographs.ParticipantsWe retrospectively included 66,721 fundus photographs from 3,272 eyes of 1,636 subjects to develop deep leaning models.MethodFundus photographs and visual fields were carefully examined by two independent readers from the optic disc and visual field reading centers of the ocular hypertension treatment study (OHTS). When an abnormality was detected by the readers, subject was recalled for re-testing to confirm the abnormality and further confirmation by an endpoint committee. Using OHTS data, deep learning models were trained and tested using 85% of the fundus photographs and further validated (re-tested) on the remaining (held-out) 15% of the fundus photographs.Main Outcome MeasuresAccuracy and area under the receiver-operating characteristic curve (AUC).ResultsThe AUC of the deep learning model in predicting glaucoma development 4-7 years prior to disease onset was 0.77 (95% confidence interval 0.75, 0.79). The accuracy of the model in predicting glaucoma development about 1-3 years prior to disease onset was 0.88 (0.86, 0.91). The accuracy of the model in detecting glaucoma after onset was 0.95 (0.94, 0.96).ConclusionsDeep learning models can predict glaucoma development prior to disease onset with reasonable accuracy. Eyes with visual field abnormality but not glaucomatous optic neuropathy had a higher tendency to be missed by deep learning algorithms.

Medical Image Classification based on an Adaptive Size Deep Learning Model

2021 ◽  
Vol 17 (3s) ◽  
pp. 1-18
Author(s):  
Xiangbin Liu ◽  
Jiesheng He ◽  
Liping Song ◽  
Shuai Liu ◽  
Gautam Srivastava
Keyword(s):  
Deep Learning ◽  
Image Classification ◽  
Medical Image ◽  
Rapid Development ◽  
Learning Model ◽  
Fine Tuning ◽  
Optimal Size ◽  
Image Dataset ◽  
Medical Image Classification ◽  
Deep Learning Model

With the rapid development of Artificial Intelligence (AI), deep learning has increasingly become a research hotspot in various fields, such as medical image classification. Traditional deep learning models use Bilinear Interpolation when processing classification tasks of multi-size medical image dataset, which will cause the loss of information of the image, and then affect the classification effect. In response to this problem, this work proposes a solution for an adaptive size deep learning model. First, according to the characteristics of the multi-size medical image dataset, the optimal size set module is proposed in combination with the unpooling process. Next, an adaptive deep learning model module is proposed based on the existing deep learning model. Then, the model is fused with the size fine-tuning module used to process multi-size medical images to obtain a solution of the adaptive size deep learning model. Finally, the proposed solution model is applied to the pneumonia CT medical image dataset. Through experiments, it can be seen that the model has strong robustness, and the classification effect is improved by about 4% compared with traditional algorithms.

Extracting Grain Orientations from EBSD Patterns of Polycrystalline Materials Using Convolutional Neural Networks

2018 ◽  
Vol 24 (5) ◽  
pp. 497-502 ◽  
Author(s):  
Dipendra Jha ◽  
Saransh Singh ◽  
Reda Al-Bahrani ◽  
Wei-keng Liao ◽  
Alok Choudhary ◽  
...  
Keyword(s):  
Deep Learning ◽  
Network Architecture ◽  
Crystal Orientation ◽  
Electron Backscatter Diffraction ◽  
Ground Truth ◽  
Learning Model ◽  
Stochastic Gradient Descent ◽  
Polycrystalline Materials ◽  
Polycrystalline Nickel ◽  
Deep Learning Model

AbstractWe present a deep learning approach to the indexing of electron backscatter diffraction (EBSD) patterns. We design and implement a deep convolutional neural network architecture to predict crystal orientation from the EBSD patterns. We design a differentiable approximation to the disorientation function between the predicted crystal orientation and the ground truth; the deep learning model optimizes for the mean disorientation error between the predicted crystal orientation and the ground truth using stochastic gradient descent. The deep learning model is trained using 374,852 EBSD patterns of polycrystalline nickel from simulation and evaluated using 1,000 experimental EBSD patterns of polycrystalline nickel. The deep learning model results in a mean disorientation error of 0.548° compared to 0.652° using dictionary based indexing.

scholarly journals Do Game Data Generalize Well for Remote Sensing Image Segmentation?

2020 ◽  
Vol 12 (2) ◽  
pp. 275 ◽  
Author(s):  
Zhengxia Zou ◽  
Tianyang Shi ◽  
Wenyuan Li ◽  
Zhou Zhang ◽  
Zhenwei Shi
Keyword(s):  
Remote Sensing ◽  
Deep Learning ◽  
Video Game ◽  
Real World ◽  
Remote Sensing Data ◽  
Ground Truth ◽  
Remote Sensing Image ◽  
Learning Model ◽  
Image Translation ◽  
Deep Learning Model

Despite the recent progress in deep learning and remote sensing image interpretation, the adaption of a deep learning model between different sources of remote sensing data still remains a challenge. This paper investigates an interesting question: do synthetic data generalize well for remote sensing image applications? To answer this question, we take the building segmentation as an example by training a deep learning model on the city map of a well-known video game “Grand Theft Auto V” and then adapting the model to real-world remote sensing images. We propose a generative adversarial training based segmentation framework to improve the adaptability of the segmentation model. Our model consists of a CycleGAN model and a ResNet based segmentation network, where the former one is a well-known image-to-image translation framework which learns a mapping of the image from the game domain to the remote sensing domain; and the latter one learns to predict pixel-wise building masks based on the transformed data. All models in our method can be trained in an end-to-end fashion. The segmentation model can be trained without using any additional ground truth reference of the real-world images. Experimental results on a public building segmentation dataset suggest the effectiveness of our adaptation method. Our method shows superiority over other state-of-the-art semantic segmentation methods, for example, Deeplab-v3 and UNet. Another advantage of our method is that by introducing semantic information to the image-to-image translation framework, the image style conversion can be further improved.

Sign in / Sign up

Export Citation Format

Share Document