Histopathology-Based Deep-Learning Predicts Atherosclerotic Lesions in Intravascular Imaging

Abstract

Background: Optical coherence tomography is a powerful modality to assess atherosclerotic lesions, but detecting lesions in high-resolution OCT is challenging and requires expert knowledge. Deep-learning algorithms can be used to automatically identify atherosclerotic lesions, facilitating identification of patients at risk. We trained a deep-learning algorithm (DeepAD) with co-registered, annotated histopathology to predict atherosclerotic lesions in optical coherence tomography (OCT).Methods: Two datasets were used for training DeepAD: (i) a histopathology data set from 7 autopsy cases with 62 OCT frames and co-registered histopathology for high quality manual annotation and (ii) a clinical data set from 51 patients with 222 OCT frames in which manual annotations were based on clinical expertise only. A U-net based deep convolutional neural network (CNN) ensemble was employed as an atherosclerotic lesion prediction algorithm. Results were analyzed using intersection over union (IOU) for segmentation.Results: DeepAD showed good performance regarding the prediction of atherosclerotic lesions, with a median IOU of 0.68 ± 0.18 for segmentation of atherosclerotic lesions. Detection of calcified lesions yielded an IOU = 0.34. When training the algorithm without histopathology-based annotations, a performance drop of >0.25 IOU was observed. The practical application of DeepAD was evaluated retrospectively in a clinical cohort (n = 11 cases), showing high sensitivity as well as specificity and similar performance when compared to manual expert analysis.Conclusion: Automated detection of atherosclerotic lesions in OCT is improved using a histopathology-based deep-learning algorithm, allowing accurate detection in the clinical setting. An automated decision-support tool based on DeepAD could help in risk prediction and guide interventional treatment decisions.

DeepNeo: Deep Learning for neointimal tissue characterization using optical coherence tomography

“…With an aging population requiring medical attention, use of deep learning-based algorithms for clinical decision support and hence reduced workload is reasonable and has already been demonstrated in different fields of medicine [33, 34]. Previous works have demonstrated the ability to segment and characterize native atherosclerotic lesions using artificial intelligence-enhanced OCT [16, 17, 35, 36]. However, to the best of our knowledge, no study so far has investigated the potential of deep learning to facilitate OCT-based characterization of neointima.…”

“…We believe that DeepNeo, which allows quick and intuitive, fully-automated characterization of the underlying neointima without requiring additional human input, would be useful in following up on vulnerable patients. DeepNeo, in combination with DeepAD [17], our previously published work on the detection of native atherosclerotic lesions, provides interventional cardiologists with a useful toolbox for facilitating OCT interpretation on native as well as stented segments.…”

“…Deep learning has the potential to greatly assist clinicians in accurately diagnosing patients through the analysis of medical images [14, 15]. In intravascular OCT imaging, deep learning has been successfully used to characterize native atherosclerotic lesions [16, 17]. In this study, we present the first fully-automated deep learning-based algorithm (“DeepNeo”) that enables quick and accurate automated segmentation and classification of neointimal tissue characteristics.…”

Automated Segmentation of Intracranial Carotid Atherosclerosis in Histological Images: Assessing the Effect of Staining

“…Although traditional computer vision methods have been successfully used for WSI analysis 20,21 , deep convolution neural networks (CNN) generally achieve better results [22][23][24] . In recent years, several architectures based on different CNN architectures have been applied to the task of WSI segmentation, such as U-Net [25][26][27][28] , v3_DCNN 29,30 , Inception V3 29,31 , DeepLabV3 + 24,32,33 , HistoCEA 34 , HistoSegNet 35 , ResNet 36,37 , etc. We chose the U-Net and DeepLabV3 + models due to their popularity and effectiveness in WSI segmentation.…”