Diagnostic Assessment of Deep Learning Algorithms for Frozen Tissue Section Analysis in Women with Breast Cancer.

PURPOSE: Assessing the metastasis status of the sentinel lymph nodes (SLNs) for hematoxylin and eosin-stained frozen tissue sections by pathologists is an essential but tedious and time-consuming task that contributes to accurate breast cancer staging. This study aimed to review a challenge competition (HeLP 2019) for the development of automated solutions for classifying the metastasis status of breast cancer patients. Materials and Methods: A total of 524 digital slides were obtained from frozen SLN sections: 297 (56.7%) from Asan Medical Center (AMC) and 227 (43.4%) from Seoul National University Bundang Hospital (SNUBH), South Korea. The slides were divided into training, development, and validation sets, where the development set comprised slides from both institutions and training and validation set included slides from only AMC and SNUBH, respectively. The algorithms were assessed for area under the receiver operating characteristic curve (AUC) and measurement of the longest metastatic tumor diameter. The final total scores were calculated as the mean of the two metrics, and the three teams with AUC values greater than 0.500 were selected for review and analysis in this study. RESULTS: The top three teams showed AUC values of 0.891, 0.809, and 0.736 and major axis prediction scores of 0.525, 0.459, and 0.387 for the validation set. The major factor that lowered the diagnostic accuracy was micro-metastasis. CONCLUSION: In this challenge competition, accurate deep learning algorithms were developed that can be helpful for making a diagnosis on intraoperative SLN biopsy. The clinical utility of this approach was evaluated by including an external validation set from SNUBH.

Diffusion and perfusion MRI radiomics obtained from deep learning segmentation provides reproducible and comparable diagnostic model to human in post-treatment glioblastoma.

OBJECTIVES: Deep learning-based automatic segmentation (DLAS) helps the reproducibility of radiomics features, but its effect on radiomics modeling is unknown. We therefore evaluated whether DLAS can robustly extract anatomical and physiological MRI features, thereby assisting in the accurate assessment of treatment response in glioblastoma patients. METHODS: A DLAS model was trained on 238 glioblastomas and validated on an independent set of 98 pre- and 86 post-treatment glioblastomas from two tertiary hospitals. A total of 1618 radiomics features from contrast-enhanced T1-weighted images (CE-T1w) and histogram features from apparent diffusion coefficient (ADC) and cerebral blood volume (CBV) mapping were extracted. The diagnostic performance of radiomics features and ADC and CBV parameters for identifying treatment response was tested using area under the curve (AUC) from receiver operating characteristics analysis. Feature reproducibility was tested using a 0.80 cutoff for concordance correlation coefficients. RESULTS: Reproducibility was excellent for ADC and CBV features (ICC, 0.82-0.99) and first-order features (pre- and post-treatment, 100% and 94.1% remained), but lower for texture (79.0% and 69.1% remained) and wavelet-transformed (81.8% and 74.9% remained) features of CE-T1w. DLAS-based radiomics showed similar performance to human-performed segmentations in internal validation (AUC, 0.81 [95% CI, 0.64-0.99] vs. AUC, 0.81 [0.60-1.00], p = 0.80), but slightly lower performance in external validation (AUC, 0.78 [0.61-0.95] vs. AUC, 0.65 [0.46-0.84], p = 0.23). CONCLUSION: DLAS-based feature extraction showed high reproducibility for first-order features from anatomical and physiological MRI, and comparable diagnostic performance to human manual segmentations in the identification of pseudoprogression, supporting the utility of DLAS in quantitative MRI analysis. KEY POINTS: • Deep learning-based automatic segmentation (DLAS) enables fast and robust feature extraction from diffusion- and perfusion-weighted MRI. • DLAS showed high reproducibility in first-order feature extraction from anatomical, diffusion, and perfusion MRI across two centers. • DLAS-based radiomics features showed comparable diagnostic accuracy to manual segmentations in post-treatment glioblastoma.

Challenge for Diagnostic Assessment of Deep Learning Algorithm for Metastases Classification in Sentinel Lymph Nodes on Frozen Tissue Section Digital Slides in Women with Breast Cancer.

PURPOSE: Assessing the status of metastasis in sentinel lymph nodes (SLNs) by pathologists is an essential task for the accurate staging of breast cancer. However, histopathological evaluation of SLNs by a pathologist is not easy and is a tedious and time-consuming task. The purpose of this study is to review a challenge competition (HeLP 2018) to develop automated solutions for the classification of metastases in hematoxylin and eosin-stained frozen tissue sections of SLNs in breast cancer patients. MATERIALS AND METHODS: A total of 297 digital slides were obtained from frozen SLN sections, which include post-neoadjuvant cases (n = 144, 48.5%) in Asan Medical Center, South Korea. The slides were divided into training, development, and validation sets. All of the imaging datasets have been manually segmented by expert pathologists. A total of 10 participants were allowed to use the Kakao challenge platform for six weeks with two P40 GPUs. The algorithms were assessed in terms of the AUC (area under receiver operating characteristic curve). RESULTS: The top three teams showed 0.986, 0.985, and 0.945 AUCs for the development set and 0.805, 0.776, and 0.765 AUCs for the validation set. Micrometastatic tumors, neoadjuvant systemic therapy, invasive lobular carcinoma, and histologic grade 3 were associated with lower diagnostic accuracy. CONCLUSION: In a challenge competition, accurate deep learning algorithms have been developed, which can be helpful in making frozen diagnosis of intraoperative SLN biopsy. Whether this approach has clinical utility will require evaluation in a clinical setting.

Automated detection algorithm for C4d immunostaining showed comparable diagnostic performance to pathologists in renal allograft biopsy.

A deep learning-based image analysis could improve diagnostic accuracy and efficiency in pathology work. Recently, we proposed a deep learning-based detection algorithm for C4d immunostaining in renal allografts. The objective of this study is to assess the diagnostic performance of the algorithm by comparing pathologists' diagnoses and analyzing the associations of the algorithm with clinical data. C4d immunostaining slides of renal allografts were obtained from two different institutions (100 slides from the Asan Medical Center and 86 slides from the Seoul National University Hospital) and scanned using two different slide scanners. Three pathologists and the algorithm independently evaluated each slide according to the Banff 2017 criteria. Subsequently, they jointly reviewed the results for consensus scoring. The result of the algorithm was compared with that of each pathologist and the consensus diagnosis. Clinicopathological associations of the results of the algorithm with allograft survival, histologic evidence of microvascular inflammation, and serologic results for donor-specific antibodies were also analyzed. As a result, the reproducibility between the pathologists was fair to moderate (kappa 0.36-0.54), which is comparable to that between the algorithm and each pathologist (kappa 0.34-0.51). The C4d scores predicted by the algorithm achieved substantial concordance with the consensus diagnosis (kappa = 0.61), and they were significantly associated with remarkable microvascular inflammation (P = 0.001), higher detection rate of donor-specific antibody (P = 0.003), and shorter graft survival (P < 0.001). In conclusion, the deep learning-based C4d detection algorithm showed a diagnostic performance similar to that of the pathologists.

Radiomic features and multilayer perceptron network classifier: a robust MRI classification strategy for distinguishing glioblastoma from primary central nervous system lymphoma.

We aimed to establish a high-performing and robust classification strategy, using magnetic resonance imaging (MRI), along with combinations of feature extraction and selection in human and machine learning using radiomics or deep features by employing a small dataset. Using diffusion and contrast-enhanced T1-weighted MR images obtained from patients with glioblastomas and primary central nervous system lymphomas, classification task was assigned to a combination of radiomic features and (1) supervised machine learning after feature selection or (2) multilayer perceptron (MLP) network; or MR image input without radiomic feature extraction to (3) two neuro-radiologists or (4) an end-to-end convolutional neural network (CNN). The results showed similar high performance in generalized linear model (GLM) classifier and MLP using radiomics features in the internal validation set, but MLP network remained robust in the external validation set obtained using different MRI protocols. CNN showed the lowest performance in both validation sets. Our results reveal that a combination of radiomic features and MLP network classifier serves a high-performing and generalizable model for classification task for a small dataset with heterogeneous MRI protocols.

Interactive registration between supine and prone scans in computed tomography colonography using band-height images.

In computed tomographic colonography (CTC), a patient is commonly scanned twice including supine and prone scans to improve the sensitivity of polyp detection. Typically, a radiologist must manually match the corresponding areas in the supine and prone CT scans, which is a difficult and time-consuming task, even for experienced scan readers. In this paper, we propose a method of supine-prone registration utilizing band-height images, which are directly constructed from the CT scans using a ray-casting algorithm containing neighboring shape information. In our method, we first identify anatomical feature points and establish initial correspondences using local extreme points on centerlines. We then correct correspondences using band-height images that contain neighboring shape information. We use geometrical and image-based information to match positions between the supine and prone centerlines. Finally, our algorithm searches the correspondence of user input points using the matched anatomical feature point pairs as key points and band-height images. The proposed method achieved accurate matching and relatively faster processing time than other previously proposed methods. The mean error of the matching between the supine and prone points for uniformly sampled positions was 18.41±22.07mm in 20 CTC datasets. The average pre-processing time was 62.9±8.6s, and the interactive matching was performed in nearly real-time. Our supine-prone registration method is expected to be helpful for the accurate and fast diagnosis of polyps.

Fast three-material modeling with triple arch projection for electronic cleansing in CTC.

In this paper, we propose a fast three-material modeling for electronic cleansing (EC) in computed tomographic colonography. Using a triple arch projection, our three-material modeling provides a very quick estimate of the three-material fractions to remove ridge-shaped artifacts at the T-junctions where air, soft-tissue (ST), and tagged residues (TRs) meet simultaneously. In our approach, colonic components including air, TR, the layer between air and TR, the layer between ST and TR (L(ST/TR)), and the T-junction are first segmented. Subsequently, the material fraction of ST for each voxel in L(ST/TR) and the T-junction is determined. Two-material fractions of the voxels in L(ST/TR) are derived based on a two-material transition model. On the other hand, three-material fractions of the voxels in the T-junction are estimated based on our fast three-material modeling with triple arch projection. Finally, the CT density value of each voxel is updated based on our fold-preserving reconstruction model. Experimental results using ten clinical datasets demonstrate that the proposed three-material modeling successfully removed the T-junction artifacts and clearly reconstructed the whole colon surface while preserving the submerged folds well. Furthermore, compared with the previous three-material transition model, the proposed three-material modeling resulted in about a five-fold increase in speed with the better preservation of submerged folds and the similar level of cleansing quality in T-junction regions.

EBNA1 binding and epigenetic regulation of gastrokine tumor suppressor genes in gastric carcinoma cells.

BACKGROUND: Epstein-Barr Virus (EBV) latently infects ~10% of gastric carcinomas (GC). Epstein-Barr Nuclear Antigen 1 (EBNA1) is expressed in EBV-associated GC, and can bind host DNA, where it may impact cellular gene regulation. Here, we show that EBNA1 binds directly to DNA upstream of the divergently transcribed GC-specific tumor suppressor genes gastrokine 1 (GKN1) and gastrokine 2 (GKN2). METHODS: We use ChIP-Seq, ChIP-qPCR, and EMSA to demonstrate that EBNA1 binds directly to the GKN1 and GKN2 promoter locus. We generate AGS-EBV, and AGS-EBNA1 cell lines to study the effects of EBNA1 on GKN1 and GKN2 mRNA expression with or without 5' azacytidine treatment. RESULTS: We show that gastrokine genes are transcriptionally silenced by DNA methylation. We also show that latent EBV infection further reduces GKN1 and GKN2 expression in AGS gastric carcinoma cells, and that siRNA depletion of EBNA1 partially alleviates this repression. However, ectopic expression of EBNA1 slightly increased GKN1 and GKN2 basal mRNA levels, but reduced their responsiveness to demethylating agent. CONCLUSIONS: These findings demonstrate that EBNA1 binds to the divergent promoter of the GKN1 and GKN2 genes in GC cells, and suggest that EBNA1 contributes to the complex transcriptional and epigenetic deregulation of the GKN1 and GKN2 tumor suppressor genes in EBV positive GC.

Fold-preserving electronic cleansing using a reconstruction model integrating material fractions and structural responses.

In this paper, we propose an electronic cleansing method using a novel reconstruction model for removing tagged materials (TMs) in computed tomography (CT) images. To address the partial volume (PV) and pseudoenhancement (PEH) effects concurrently, material fractions and structural responses are integrated into a single reconstruction model. In our approach, colonic components including air, TM, an interface layer between air and TM, and an interface layer between soft-tissue (ST) and TM (IL ST/TM ) are first segmented. For each voxel in IL ST/TM, the material fractions of ST and TM are derived using a two-material transition model, and the structural response to identify the folds submerged in the TM is calculated by the rut-enhancement function based on the eigenvalue signatures of the Hessian matrix. Then, the CT density value of each voxel in IL ST/TM is reconstructed based on both the material fractions and structural responses. The material fractions remove the aliasing artifacts caused by a PV effect in IL ST/TM effectively while the structural responses avoid the erroneous cleansing of the submerged folds caused by the PEH effect. Experimental results using ten clinical datasets demonstrated that the proposed method showed higher cleansing quality and better preservation of submerged folds than the previous method, which was validated by the higher mean density values and fold preservation rates for manually segmented fold regions.

Does the amount of tagged stool and fluid significantly affect the radiation exposure in low-dose CT colonography performed with an automatic exposure control?

OBJECTIVE: To determine whether the amount of tagged stool and fluid significantly affects the radiation exposure in low-dose screening CT colonography performed with an automatic tube-current modulation technique. METHODS: The study included 311 patients. The tagging agent was barium (n = 271) or iodine (n = 40). Correlation was measured between mean volume CT dose index (CTDI (vol)) and the estimated x-ray attenuation of the tagged stool and fluid (ATT). Multiple linear regression analyses were performed to determine the effect of ATT on CTDI (vol ) and the effect of ATT on image noise while adjusting for other variables including abdominal circumference. RESULTS: CTDI (vol) varied from 0.88 to 2.54 mGy. There was no significant correlation between CTDI (vol) and ATT (p = 0.61). ATT did not significantly affect CTDI (vol) (p = 0.93), while abdominal circumference was the only factor significantly affecting CTDI (vol) (p < 0.001). Image noise ranged from 59.5 to 64.1 HU. The p value for the regression model explaining the noise was 0.38. CONCLUSION: The amount of stool and fluid tagging does not significantly affect radiation exposure.