Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759563
T. L. Koch, Mathias Perslev, C. Igel, Sami Sebastian Brandt
Fully convolutional neural networks (FCNs) have proven to be powerful tools for medical image segmentation. We apply an FCN based on the U-Net architecture for the challenging task of semantic segmentation of dental panoramic radiographs and discuss general tricks for improving segmentation performance. Among those are network ensembling, test-time augmentation, data symmetry exploitation and bootstrapping of low quality annotations. The performance of our approach was tested on a highly variable dataset of 1500 dental panoramic radiographs. A single network reached the Dice score of 0.934 where 1201 images were used for training, forming an ensemble increased the score to 0.936.
{"title":"Accurate Segmentation of Dental Panoramic Radiographs with U-NETS","authors":"T. L. Koch, Mathias Perslev, C. Igel, Sami Sebastian Brandt","doi":"10.1109/ISBI.2019.8759563","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759563","url":null,"abstract":"Fully convolutional neural networks (FCNs) have proven to be powerful tools for medical image segmentation. We apply an FCN based on the U-Net architecture for the challenging task of semantic segmentation of dental panoramic radiographs and discuss general tricks for improving segmentation performance. Among those are network ensembling, test-time augmentation, data symmetry exploitation and bootstrapping of low quality annotations. The performance of our approach was tested on a highly variable dataset of 1500 dental panoramic radiographs. A single network reached the Dice score of 0.934 where 1201 images were used for training, forming an ensemble increased the score to 0.936.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128270872","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759453
Hongjian Wang, T. Huynh, H. Gemmeke, T. Hopp, J. Hesser
To accelerate the process of 3D ultrasound computed tomography, we parallelize the most time-consuming part of a paraxial forward model on GPU, where massive complex multiplications and 2D Fourier transforms have to be performed iteratively. We test our GPU implementation on a synthesized symmetric breast phantom with different sizes. In the best case, for only one emitter position, the speedup of a desktop GPU reaches 23 times when the data transfer time is included, and 100 times when only GPU parallel computing time is considered. In the worst case, the speedup of a less powerful laptop GPU is still 2.5 times over a six-core desktop CPU, when the data transfer time is included. For the correctness of the values computed on GPU, the maximum percent deviation of L2 norm is only 0.014%.
{"title":"GPU Acceleration of Wave Based Transmission Tomography","authors":"Hongjian Wang, T. Huynh, H. Gemmeke, T. Hopp, J. Hesser","doi":"10.1109/ISBI.2019.8759453","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759453","url":null,"abstract":"To accelerate the process of 3D ultrasound computed tomography, we parallelize the most time-consuming part of a paraxial forward model on GPU, where massive complex multiplications and 2D Fourier transforms have to be performed iteratively. We test our GPU implementation on a synthesized symmetric breast phantom with different sizes. In the best case, for only one emitter position, the speedup of a desktop GPU reaches 23 times when the data transfer time is included, and 100 times when only GPU parallel computing time is considered. In the worst case, the speedup of a less powerful laptop GPU is still 2.5 times over a six-core desktop CPU, when the data transfer time is included. For the correctness of the values computed on GPU, the maximum percent deviation of L2 norm is only 0.014%.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130310842","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759435
Li Liu, Da Chen, L. Cohen, H. Shu, M. Pâques
In this work, we propose a new minimal path model with a dynamic Riemannian metric to overcome the shortcuts problem in vessel extraction. The invoked metric consists of a crossing-adaptive anisotropic radius-lifted tensor field and a front freezing indicator. It is able to reduce the anisotropy of the metric on the crossing points and steer the front evolution by freezing the points causing high curvature of a geodesic. We validate our model on the DRIVE and IOSTAR datasets, and the segmentation accuracy is 0.861 and 0.881, respectively. The proposed method can extract the centreline position and vessel width efficiently and accuracy.
{"title":"Vessel Extraction Using Crossing-Adaptive Minimal Path Model With Anisotropic Enhancement And Curvature Constraint","authors":"Li Liu, Da Chen, L. Cohen, H. Shu, M. Pâques","doi":"10.1109/ISBI.2019.8759435","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759435","url":null,"abstract":"In this work, we propose a new minimal path model with a dynamic Riemannian metric to overcome the shortcuts problem in vessel extraction. The invoked metric consists of a crossing-adaptive anisotropic radius-lifted tensor field and a front freezing indicator. It is able to reduce the anisotropy of the metric on the crossing points and steer the front evolution by freezing the points causing high curvature of a geodesic. We validate our model on the DRIVE and IOSTAR datasets, and the segmentation accuracy is 0.861 and 0.881, respectively. The proposed method can extract the centreline position and vessel width efficiently and accuracy.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"119 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130804881","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759292
M. Jafari, H. Girgis, A. Abdi, Zhibin Liao, Mehran Pesteie, R. Rohling, K. Gin, T. Tsang, P. Abolmaesumi
Accurate segmentation of left ventricle (LV) in apical four chamber echocardiography cine is a key step in cardiac functionality assessment. Cardiologists roughly annotate two frames in the cardiac cycle, namely, the end-diastolic and end-systolic frames, as part of their clinical workflow, limiting the annotated data to less than 5% of the frames in the cardiac cycle. In this paper, we propose a semi-supervised learning algorithm to leverage the unlabeled data to improve the performance of LV segmentation algorithms. This approach is based on a generative model which learns an inverse mapping from segmentation masks to their corresponding echo frames. This generator is then used as a critic to assess and improve the LV segmentation mask generated by a given segmentation algorithm such as U-Net. This semi-supervised approach enforces a prior on the segmentation model based on the perceptual similarity of the generated frame with the original frame. This approach promotes utilization of the unlabeled samples, which, in turn, improves the segmentation accuracy.
{"title":"Semi-Supervised Learning For Cardiac Left Ventricle Segmentation Using Conditional Deep Generative Models as Prior","authors":"M. Jafari, H. Girgis, A. Abdi, Zhibin Liao, Mehran Pesteie, R. Rohling, K. Gin, T. Tsang, P. Abolmaesumi","doi":"10.1109/ISBI.2019.8759292","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759292","url":null,"abstract":"Accurate segmentation of left ventricle (LV) in apical four chamber echocardiography cine is a key step in cardiac functionality assessment. Cardiologists roughly annotate two frames in the cardiac cycle, namely, the end-diastolic and end-systolic frames, as part of their clinical workflow, limiting the annotated data to less than 5% of the frames in the cardiac cycle. In this paper, we propose a semi-supervised learning algorithm to leverage the unlabeled data to improve the performance of LV segmentation algorithms. This approach is based on a generative model which learns an inverse mapping from segmentation masks to their corresponding echo frames. This generator is then used as a critic to assess and improve the LV segmentation mask generated by a given segmentation algorithm such as U-Net. This semi-supervised approach enforces a prior on the segmentation model based on the perceptual similarity of the generated frame with the original frame. This approach promotes utilization of the unlabeled samples, which, in turn, improves the segmentation accuracy.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129999353","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759173
Zeya Wang, Nanqing Dong, Sean D. Rosario, Min Xu, P. Xie, E. Xing
Glaucoma is an eye disease that damages the optic nerve and leads to loss of vision. The diagnosis of glaucoma involves measurement of cup-to-disc ratio from retinal fundus images, which necessitates the detection of the optic disc-and-cup boundary as a crucial task for glaucoma screening. Most existing computer-aided diagnosis (CAD) systems focus on the segmentation approaches but ignore the localization approaches, which requires less human annotation cost. In this paper, we propose a deep learning-based framework to jointly localize the ellipse for the optic disc (OD) and optic cup (OC) regions. Instead of detecting a bounding box like in most object detection approaches, we directly estimate the parameters of an ellipse that suffices to capture the morphology of each OD and OC region for calculating the cup-to-disc ratio. We use two modules to detect the ellipses for OD and OC regions, where the OD region serves as attention to the OC region. The proposed framework achieves competitive results against the state-of-the-art segmentation methods with less supervision. We empirically evaluate our framework with the recent state-of-the-art segmentation models on two scenarios where the training data and test data come from the same and different domains.
{"title":"Ellipse Detection of Optic Disc-and-Cup Boundary in Fundus Images","authors":"Zeya Wang, Nanqing Dong, Sean D. Rosario, Min Xu, P. Xie, E. Xing","doi":"10.1109/ISBI.2019.8759173","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759173","url":null,"abstract":"Glaucoma is an eye disease that damages the optic nerve and leads to loss of vision. The diagnosis of glaucoma involves measurement of cup-to-disc ratio from retinal fundus images, which necessitates the detection of the optic disc-and-cup boundary as a crucial task for glaucoma screening. Most existing computer-aided diagnosis (CAD) systems focus on the segmentation approaches but ignore the localization approaches, which requires less human annotation cost. In this paper, we propose a deep learning-based framework to jointly localize the ellipse for the optic disc (OD) and optic cup (OC) regions. Instead of detecting a bounding box like in most object detection approaches, we directly estimate the parameters of an ellipse that suffices to capture the morphology of each OD and OC region for calculating the cup-to-disc ratio. We use two modules to detect the ellipses for OD and OC regions, where the OD region serves as attention to the OC region. The proposed framework achieves competitive results against the state-of-the-art segmentation methods with less supervision. We empirically evaluate our framework with the recent state-of-the-art segmentation models on two scenarios where the training data and test data come from the same and different domains.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125571963","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759571
F. G. Zanjani, Andreas Panteli, S. Zinger, F. V. D. Sommen, T. Tan, Benjamin Balluff, D. Vos, S. Ellis, R. Heeren, M. Lucas, H. Marquering, Ivo G. H. Jansen, C. D. Savci-Heijink, D. M. Bruin, P. D. With
Mass spectrometry imaging (MSI) reveals the localization of a broad scale of compounds ranging from metabolites to proteins in biological tissues. This makes MSI an attractive tool in biomedical research for studying diseases. Computer-aided diagnosis (CAD) systems facilitate the analysis of the molecular profile in tumor tissues to provide a distinctive fingerprint for finding biomarkers. In this paper, the performance of recurrent neural networks (RNNs) is studied on MSI data to exploit their learning capabilities for finding irregular patterns and dependencies in sequential data. In order to design a better CAD model for tumor detection/classification, several configurations of Long Short-Time Memory (LSTM) are examined. The proposed model consists of a 2-layer bidirectional LSTM, each containing 100 LSTM units. The proposed RNN model outperforms the state-of-the-art CNN model by 1.87% and 1.45% higher accuracy in mass spectra classification on lung and bladder cancer datasets with a sixfold faster training time.
{"title":"Cancer Detection in Mass Spectrometry Imaging Data by Recurrent Neural Networks","authors":"F. G. Zanjani, Andreas Panteli, S. Zinger, F. V. D. Sommen, T. Tan, Benjamin Balluff, D. Vos, S. Ellis, R. Heeren, M. Lucas, H. Marquering, Ivo G. H. Jansen, C. D. Savci-Heijink, D. M. Bruin, P. D. With","doi":"10.1109/ISBI.2019.8759571","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759571","url":null,"abstract":"Mass spectrometry imaging (MSI) reveals the localization of a broad scale of compounds ranging from metabolites to proteins in biological tissues. This makes MSI an attractive tool in biomedical research for studying diseases. Computer-aided diagnosis (CAD) systems facilitate the analysis of the molecular profile in tumor tissues to provide a distinctive fingerprint for finding biomarkers. In this paper, the performance of recurrent neural networks (RNNs) is studied on MSI data to exploit their learning capabilities for finding irregular patterns and dependencies in sequential data. In order to design a better CAD model for tumor detection/classification, several configurations of Long Short-Time Memory (LSTM) are examined. The proposed model consists of a 2-layer bidirectional LSTM, each containing 100 LSTM units. The proposed RNN model outperforms the state-of-the-art CNN model by 1.87% and 1.45% higher accuracy in mass spectra classification on lung and bladder cancer datasets with a sixfold faster training time.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125755778","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759575
Kristofer Pomiecko, Carson D. Sestili, K. Fissell, S. Pathak, D. Okonkwo, W. Schneider
This paper presents an application of 3D convolutional neural network (CNN) techniques to compute the white matter region spanned by a fiber tract (the tract mask) from whole-brain MRI diffusion anisotropy maps. The DeepMedic CNN platform was used, allowing for training directly on 3D volumes. The dataset consisted of 240 subjects, controls and traumatic brain injury (TBI) patients, scanned with a high angular direction and high b-value multi-shell diffusion protocol. Twelve tract masks per subject were learned. Median Dice scores of 0.72 were achieved over the 720 test masks in comparing learned tract masks to manually created masks. This work demonstrates ability to learn complex spatial regions in control and patient populations and contributes a new application of CNNs as a fast pre-selection tool in automated white matter tract segmentation methods.
{"title":"3D Convolutional Neural Network Segmentation of White Matter Tract Masks from MR Diffusion Anisotropy Maps","authors":"Kristofer Pomiecko, Carson D. Sestili, K. Fissell, S. Pathak, D. Okonkwo, W. Schneider","doi":"10.1109/ISBI.2019.8759575","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759575","url":null,"abstract":"This paper presents an application of 3D convolutional neural network (CNN) techniques to compute the white matter region spanned by a fiber tract (the tract mask) from whole-brain MRI diffusion anisotropy maps. The DeepMedic CNN platform was used, allowing for training directly on 3D volumes. The dataset consisted of 240 subjects, controls and traumatic brain injury (TBI) patients, scanned with a high angular direction and high b-value multi-shell diffusion protocol. Twelve tract masks per subject were learned. Median Dice scores of 0.72 were achieved over the 720 test masks in comparing learned tract masks to manually created masks. This work demonstrates ability to learn complex spatial regions in control and patient populations and contributes a new application of CNNs as a fast pre-selection tool in automated white matter tract segmentation methods.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133981259","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759235
Rémy Dubois, Arthur Imbert, Aubin Samacoïts, M. Peter, E. Bertrand, Florian Müller, Thomas Walter
The localization of messenger RNA (mRNA) molecules inside cells play an important role for the local control of gene expression. However, the localization patterns of many mRNAs remain unknown and poorly understood. Single Molecule Fluorescence in Situ Hybridization (smFISH) allows for the visualization of individual mRNA molecules in cells. This method is now scalable and can be applied in High Content Screening (HCS) mode. Here, we propose a computational workflow based on deep convolutional neural networks trained on simulated data to identify different localization patterns from large-scale smFISH data.
{"title":"A Deep Learning Approach To Identify MRNA Localization Patterns","authors":"Rémy Dubois, Arthur Imbert, Aubin Samacoïts, M. Peter, E. Bertrand, Florian Müller, Thomas Walter","doi":"10.1109/ISBI.2019.8759235","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759235","url":null,"abstract":"The localization of messenger RNA (mRNA) molecules inside cells play an important role for the local control of gene expression. However, the localization patterns of many mRNAs remain unknown and poorly understood. Single Molecule Fluorescence in Situ Hybridization (smFISH) allows for the visualization of individual mRNA molecules in cells. This method is now scalable and can be applied in High Content Screening (HCS) mode. Here, we propose a computational workflow based on deep convolutional neural networks trained on simulated data to identify different localization patterns from large-scale smFISH data.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"7 2","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"113964475","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759291
Abdullah Nazib, C. Fookes, Dimitri Perrin
Image registration plays an important role in comparing images. It is particularly important in analysing medical images like CT, MRI and PET, to quantify different biological samples, to monitor disease progression, and to fuse different modalities to support better diagnosis. The recent emergence of tissue clearing protocols enable us to take images at cellular level resolution. Image registration tools developed for other modalities are currently unable to manage images of entire organs at such resolution. The popularity of deep learning based methods in the computer vision community justifies a rigorous investigation of deep-learning based methods on tissue cleared images along with their traditional counterparts. In this paper, we investigate and compare the performance of a deep learning based registration method with traditional optimization based methods on samples from tissue-clearing methods. From the comparative results it is found that a deep-learning based method outperforms all traditional registration tools in terms of registration time and has achieved promising registration accuracy.
{"title":"Towards Extreme-Resolution Image Registration with Deep Learning","authors":"Abdullah Nazib, C. Fookes, Dimitri Perrin","doi":"10.1109/ISBI.2019.8759291","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759291","url":null,"abstract":"Image registration plays an important role in comparing images. It is particularly important in analysing medical images like CT, MRI and PET, to quantify different biological samples, to monitor disease progression, and to fuse different modalities to support better diagnosis. The recent emergence of tissue clearing protocols enable us to take images at cellular level resolution. Image registration tools developed for other modalities are currently unable to manage images of entire organs at such resolution. The popularity of deep learning based methods in the computer vision community justifies a rigorous investigation of deep-learning based methods on tissue cleared images along with their traditional counterparts. In this paper, we investigate and compare the performance of a deep learning based registration method with traditional optimization based methods on samples from tissue-clearing methods. From the comparative results it is found that a deep-learning based method outperforms all traditional registration tools in terms of registration time and has achieved promising registration accuracy.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123042082","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-04-08DOI: 10.1109/ISBI.2019.8759511
A. Galdran, P. Costa, A. Campilho
Visual exploration of the larynx represents a relevant technique for the early diagnosis of laryngeal disorders. However, visualizing an endoscopy for finding abnormalities is a time-consuming process, and for this reason much research has been dedicated to the automatic analysis of endoscopic video data. In this work we address the particular task of discriminating among informative laryngoscopic frames and those that carry insufficient diagnostic information. In the latter case, the goal is also to determine the reason for this lack of information. To this end, we analyze the possibility of training three different state-of-the-art Convolutional Neural Networks, but initializing their weights from configurations that have been previously optimized for solving natural image classification problems. Our findings show that the simplest of these three architectures not only is the most accurate (outperforming previously proposed techniques), but also the fastest and most efficient, with the lowest inference time and minimal memory requirements, enabling real-time application and deployment in portable devices.
{"title":"Real-Time Informative Laryngoscopic Frame Classification with Pre-Trained Convolutional Neural Networks","authors":"A. Galdran, P. Costa, A. Campilho","doi":"10.1109/ISBI.2019.8759511","DOIUrl":"https://doi.org/10.1109/ISBI.2019.8759511","url":null,"abstract":"Visual exploration of the larynx represents a relevant technique for the early diagnosis of laryngeal disorders. However, visualizing an endoscopy for finding abnormalities is a time-consuming process, and for this reason much research has been dedicated to the automatic analysis of endoscopic video data. In this work we address the particular task of discriminating among informative laryngoscopic frames and those that carry insufficient diagnostic information. In the latter case, the goal is also to determine the reason for this lack of information. To this end, we analyze the possibility of training three different state-of-the-art Convolutional Neural Networks, but initializing their weights from configurations that have been previously optimized for solving natural image classification problems. Our findings show that the simplest of these three architectures not only is the most accurate (outperforming previously proposed techniques), but also the fastest and most efficient, with the lowest inference time and minimal memory requirements, enabling real-time application and deployment in portable devices.","PeriodicalId":119935,"journal":{"name":"2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019)","volume":"261 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114468844","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：