Deep Learning-Based Classification of Hepatocellular Nodular Lesions on Whole-Slide Histopathologic Images

Cheng, Na; Ren, Yong; Zhou, Jing; Zhang, Yiwang; Wang, Deyu; Zhang, Xiaofang; Chen, Bing; Liu, Fang; Lv, Jin; Cao, Qinghua; Chen, Sijin; Du, Hong; Dai, Hui; Weng, Zijin; Liang, Qiong; Su, Bojin; Tang, Lu-Ying; Han, Lanqing; Chen, Jianning; Shao, Chun-Kui

doi:10.1053/j.gastro.2022.02.025

Cited by 64 publications

(27 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Jin et al (10) improved this situation by developing a specially designed slide scanner that can automatically divide each WSIs into 300 patches. Cheng et al (28) invited three subspecialists to manually stretch out regions of interest (ROIs) in images, which is convenient but labor-intensive. Our study focused on the light-strategy that could acquire images from the representative biopsy tissues in a short period.…”

Section: Discussionmentioning

confidence: 99%

Predicting the recurrence and overall survival of patients with glioma based on histopathological images using deep learning

et al. 2023

View full text Add to dashboard Cite

BackgroundA deep learning (DL) model based on representative biopsy tissues can predict the recurrence and overall survival of patients with glioma, leading to optimized personalized medicine. This research aimed to develop a DL model based on hematoxylin-eosin (HE) stained pathological images and verify its diagnostic accuracy.MethodsOur study retrospectively collected 162 patients with glioma and randomly divided them into a training set (n = 113) and a validation set (n = 49) to build a DL model. The HE-stained slide was segmented into a size of 180 × 180 pixels without overlapping. The patch-level features were extracted by the pre-trained ResNet50 to predict the recurrence and overall survival. Additionally, a light-strategy was introduced where low-size digital biopsy images with clinical information were inputted into the DL model to ensure minimum memory occupation.ResultsOur study extracted 512 histopathological features from the HE-stained slides of each glioma patient. We identified 36 and 18 features as significantly related to disease-free survival (DFS) and overall survival (OS), respectively, (P < 0.05) using the univariate Cox proportional-hazards model. Pathomics signature showed a C-index of 0.630 and 0.652 for DFS and OS prediction, respectively. The time-dependent receiver operating characteristic (ROC) curves, along with nomograms, were used to assess the diagnostic accuracy at a fixed time point. In the validation set (n = 49), the area under the curve (AUC) in the 1- and 2-year DFS was 0.955 and 0.904, respectively, and the 2-, 3-, and 5-year OS were 0.969, 0.955, and 0.960, respectively. We stratified the patients into low- and high-risk groups using the median hazard score (0.083 for DFS and−0.177 for OS) and showed significant differences between these groups (P < 0.001).ConclusionOur results demonstrated that the DL model based on the HE-stained slides showed the predictability of recurrence and survival in patients with glioma. The results can be used to assist oncologists in selecting the optimal treatment strategy in clinical practice.

show abstract

Section: Discussionmentioning

confidence: 99%

Predicting the recurrence and overall survival of patients with glioma based on histopathological images using deep learning

et al. 2023

View full text Add to dashboard Cite

show abstract

“…Then three networks based on EfficientNet-B5 were trained with the three commonly used learning rates of 8e-4, 1e-3, and 1e-2 on the 161,892 patches at 40× resolution from the 127 WSIs in the training cohort, with the patients' distinct outcomes as ground truth. [30][31][32] We first chose the learning rate of 8e-4 because it is a default value configured in our medical image classification. Then, two additional learning rates of 1e-3 and 1e-2 were adopted because we aimed to see if other learning rates were helpful for performance improvement.…”

Section: Training Of Deep Learning Models For Classifications Of Trea...mentioning

confidence: 99%

Deep learning‐based prediction of treatment prognosis from nasal polyp histology slides

Wang

Ren

et al. 2022

Int Forum Allergy Rhinol

Self Cite

View full text Add to dashboard Cite

Background Histopathology of nasal polyps contains rich prognostic information, which is difficult to extract objectively. In the present study, we aimed to develop a prognostic indicator of patient outcomes by analyzing scanned conventional hematoxylin and eosin (H&E)‐stained slides alone using deep learning. Methods An interpretable supervised deep learning model was developed using 185 H&E‐stained whole‐slide images (WSIs) of nasal polyps, each from a patient randomly selected from the pool of 232 patients who underwent endoscopic sinus surgery at the First Affiliated Hospital of Sun Yat‐Sen University (internal cohort). We internally validated the model on a holdout dataset from the internal cohort (47 H&E‐stained WSIs) and externally validated the model on 122 H&E‐stained WSIs from the Seventh Affiliated Hospital of Sun Yat‐Sen University and the University of Hong Kong‐Shenzhen Hospital (external cohort). A poor prognosis score (PPS) was established to evaluate patient outcomes, and then risk activation mapping was applied to visualize the histopathological features underlying PPS. Results The model yielded a patient‐level sensitivity of 79.5%, and specificity of 92.3%, with areas under the receiver operating characteristic curve of 0.943, on the multicenter external cohort. The predictive ability of PPS was superior to that of conventional tissue eosinophil number. Notably, eosinophil infiltration, goblet cell hyperplasia, glandular hyperplasia, squamous metaplasia, and fibrin deposition were identified as the main underlying features of PPS. Conclusions Our deep learning model is an effective method for decoding pathological images of nasal polyps, providing a valuable solution for disease prognosis prediction and precise patient treatment.

show abstract

“…The abovementioned resultsprovide a glimpse into the current state of CNN applications on ECG analysis. We used 2 commonly used models VGGNet16 and ResNet50, pretrained for image classification to perform transfer learning [24,26]. The AUC values were 0.88, 0.87, and 0.76 for our CNN, and the VGGNet16, and ResNet50 models, respectively (S3 Fig) . Our CNN performs better than the ResNet50, and has almost the same performance as the VGGNet16.…”

Section: Plos Onementioning

confidence: 99%

Assessing electrocardiogram changes after ischemic stroke with artificial intelligence

Zeng

Wang

et al. 2022

PLoS ONE

View full text Add to dashboard Cite

Objective Ischemic stroke (IS) with subsequent cerebrocardiac syndrome (CCS) has a poor prognosis. We aimed to investigate electrocardiogram (ECG) changes after IS with artificial intelligence (AI). Methods We collected ECGs from a healthy population and patients with IS, and then analyzed participant demographics and ECG parameters to identify abnormal features in post-IS ECGs. Next, we trained the convolutional neural network (CNN), random forest (RF) and support vector machine (SVM) models to automatically detect the changes in the ECGs; Additionally, We compared the CNN scores of good prognosis (mRS ≤ 2) and poor prognosis (mRS > 2) to assess the prognostic value of CNN model. Finally, we used gradient class activation map (Grad-CAM) to localize the key abnormalities. Results Among the 3506 ECGs of the IS patients, 2764 ECGs (78.84%) led to an abnormal diagnosis. Then we divided ECGs in the primary cohort into three groups, normal ECGs (N-Ns), abnormal ECGs after the first ischemic stroke (A-ISs), and normal ECGs after the first ischemic stroke (N-ISs). Basic demographic and ECG parameter analyses showed that heart rate, QT interval, and P-R interval were significantly different between 673 N-ISs and 3546 N-Ns (p < 0.05). The CNN has the best performance among the three models in distinguishing A-ISs and N-Ns (AUC: 0.88, 95%CI = 0.86–0.90). The prediction scores of the A-ISs and N-ISs obtained from the all three models are statistically different from the N-Ns (p < 0.001). Futhermore, the CNN scores of the two groups (mRS > 2 and mRS ≤ 2) were significantly different (p < 0.05). Finally, Grad-CAM revealed that the V4 lead may harbor the highest probability of abnormality. Conclusion Our study showed that a high proportion of post-IS ECGs harbored abnormal changes. Our CNN model can systematically assess anomalies in and prognosticate post-IS ECGs.

show abstract

Deep Learning-Based Classification of Hepatocellular Nodular Lesions on Whole-Slide Histopathologic Images

Cited by 64 publications

References 33 publications

Predicting the recurrence and overall survival of patients with glioma based on histopathological images using deep learning

Predicting the recurrence and overall survival of patients with glioma based on histopathological images using deep learning

Deep learning‐based prediction of treatment prognosis from nasal polyp histology slides

Assessing electrocardiogram changes after ischemic stroke with artificial intelligence

Contact Info

Product

Resources

About