Deep Learning-based Recurrence Prediction in Patients with Non-muscle-invasive Bladder Cancer.

BACKGROUND: Non-muscle-invasive bladder cancer (NMIBC) is characterized by frequent recurrence of the disease, which is difficult to predict. OBJECTIVE: To combine digital histopathology slides with clinical data to predict 1- and 5-yr recurrence-free survival of NMIBC patients using deep learning. DESIGN, SETTING, AND PARTICIPANTS: Data of patients undergoing a transurethral resection of a bladder tumor between 2000 and 2018 at a Dutch academic medical center were selected. Corresponding histological slides were digitized. A three-step approach was used to predict 1- and 5-yr recurrence-free survival. First, a segmentation network was used to detect the urothelium on the digital histopathology slides. Second, a selection network was trained for the selection of patches associated with recurrence. Third, a classification network, combining the information of the selection network with clinical data, was trained to give the probability of 1- and 5-yr recurrence-free survival. OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: The accuracy of the deep learning-based model was compared with a multivariable logistic regression model using clinical data only. RESULTS AND LIMITATIONS: In the 1- and 5-yr follow-up cohorts, 359 and 281 patients were included with recurrence rates of 27% and 63%, respectively. The areas under the curve (AUCs) of the model combining digital histopathology slide data with clinical data were 0.62 and 0.76 for 1- and 5-yr recurrence predictions, respectively, which were higher than those of the model using digital histopathology slide data only (AUCs of 0.56 and 0.72, respectively) and the multivariable logistic regression (AUCs of 0.58 and 0.57, respectively). CONCLUSIONS: In our population, the deep learning-based model combining digital histopathology slides and clinical data enhances the prediction of recurrence (within 5 yr) compared with models using clinical data or image data only. PATIENT SUMMARY: By combining histopathology images and patient record data using deep learning, the prediction of recurrence in bladder cancer patients is enhanced.

Machine learning for grading and prognosis of esophageal dysplasia using mass spectrometry and histological imaging.

BACKGROUND: Barrett's esophagus (BE) is a precursor lesion of esophageal adenocarcinoma and may progress from non-dysplastic through low-grade dysplasia (LGD) to high-grade dysplasia (HGD) and cancer. Grading BE is of crucial prognostic value and is currently based on the subjective evaluation of biopsies. This study aims to investigate the potential of machine learning (ML) using spatially resolved molecular data from mass spectrometry imaging (MSI) and histological data from microscopic hematoxylin and eosin (H&E)-stained imaging for computer-aided diagnosis and prognosis of BE. METHODS: Biopsies from 57 patients were considered, divided into non-dysplastic (n = 15), LGD non-progressive (n = 14), LGD progressive (n = 14), and HGD (n = 14). MSI experiments were conducted at 50 × 50 µm spatial resolution per pixel corresponding to a tile size of 96x96 pixels in the co-registered H&E images, making a total of 144,823 tiles for the whole dataset. RESULTS: ML models were trained to distinguish epithelial tissue from stroma with area-under-the-curve (AUC) values of 0.89 (MSI) and 0.95 (H&E)) and dysplastic grade (AUC of 0.97 (MSI) and 0.85 (H&E)) on a tile level, and low-grade progressors from non-progressors on a patient level (accuracies of 0.72 (MSI) and 0.48 (H&E)). CONCLUSIONS: In summary, while the H&E-based classifier was best at distinguishing tissue types, the MSI-based model was more accurate at distinguishing dysplastic grades and patients at progression risk, which demonstrates the complementarity of both approaches. Data are available via ProteomeXchange with identifier PXD028949.

Automated Detection and Grading of Non-Muscle-Invasive Urothelial Cell Carcinoma of the Bladder.

Accurate grading of non-muscle-invasive urothelial cell carcinoma is of major importance; however, high interobserver variability exists. A fully automated detection and grading network based on deep learning is proposed to enhance reproducibility. A total of 328 transurethral resection specimens from 232 patients were included, and a consensus reading by three specialized pathologists was used. The slides were digitized, and the urothelium was annotated by expert observers. The U-Net-based segmentation network was trained to automatically detect urothelium. This detection was used as input for the classification network. The classification network aimed to grade the tumors according to the World Health Organization grading system adopted in 2004. The automated grading was compared with the consensus and individual grading. The segmentation network resulted in an accurate detection of urothelium. The automated grading shows moderate agreement (κ = 0.48 ± 0.14 SEM) with the consensus reading. The agreement among pathologists ranges between fair (κ = 0.35 ± 0.13 SEM and κ = 0.38 ± 0.11 SEM) and moderate (κ = 0.52 ± 0.13 SEM). The automated classification correctly graded 76% of the low-grade cancers and 71% of the high-grade cancers according to the consensus reading. These results indicate that deep learning can be used for the fully automated detection and grading of urothelial cell carcinoma.

Toward Automated In Vivo Bladder Tumor Stratification Using Confocal Laser Endomicroscopy.

Purpose: Urothelial carcinoma of the bladder (UCB) is the most common urinary cancer. White-light cystoscopy (WLC) forms the corner stone for the diagnosis of UCB. However, histopathological assessment is required for adjuvant treatment selection. Probe-based confocal laser endomicroscopy (pCLE) enables visualization of the microarchitecture of bladder lesions during WLC, which allows for real-time tissue differentiation and grading of UCB. To improve the diagnostic process of UCB, computer-aided classification of pCLE videos of in vivo bladder lesions were evaluated in this study. Materials and Methods: We implemented preprocessing methods to optimize contrast and to reduce striping artifacts in each individual pCLE frame. Subsequently, a semiautomatic frame selection was performed. The selected frames were used to train a feature extractor based on pretrained ImageNet networks. A recurrent neural network, in specific long short-term memory (LSTM), was used to predict the grade of bladder lesions. Differentiation of lesions was performed at two levels, namely (i) healthy and benign vs malignant tissue and (ii) low-grade vs high-grade papillary UCB. A total of 53 patients with 72 lesions were included in this study, resulting in ∼140,000 pCLE frames. Results: The semiautomated frame selection reduced the number of frames to ∼66,500 informative frames. The accuracy for differentiation of (i) healthy and benign vs malignant urothelium was 79% and (ii) high-grade and low-grade papillary UCB was 82%. Conclusions: A feature extractor in combination with LSTM results in proper stratification of pCLE videos of in vivo bladder lesions.

The First In Vivo Needle-Based Optical Coherence Tomography in Human Prostate: A Safety and Feasibility Study.

OBJECTIVE: To demonstrate the safety and feasibility of clinical in vivo needle-based optical coherence tomography (OCT) imaging of the prostate. MATERIALS AND METHODS: Two patients with prostate cancer underwent each two percutaneous in vivo needle-based OCT measurements before transperineal template mapping biopsy. The OCT probe was introduced via a needle and positioned under ultrasound guidance. To test the safety, adverse events were recorded during and after the procedure. To test the feasibility, OCT and US images were studied during and after the procedure. Corresponding regions for OCT and biopsy were determined. A uropathologist evaluated and annotated the histopathology. Three experts assessed all the corresponding OCT images. The OCT and biopsy conclusions for the corresponding regions were compared. RESULTS: No adverse events during and following the, in total four, in vivo needle-based OCT measurements were reported. The OCT measurements showed images of prostatic tissue with a penetration depth of ~1.5 mm. The histological-proven tissue types, which were also found in the overlapping OCT images, were benign glands, stroma, glandular atrophy, and adenocarcinoma (Gleason pattern 3). CONCLUSIONS: Clinical in vivo needle-based OCT of the prostate is feasible with no adverse events during measurements. OCT images displayed detailed prostatic tissue with a imaging depth up to ~1.5 mm. We could co-register four histological-proven tissue types with OCT images. The feasibility of in vivo OCT in the prostate opens the pathway to the next phase of needle-based OCT studies in the prostate. Lasers Surg. Med. 51:390-398, 2019. © 2019 The Authors. Lasers in Surgery and Medicine Published by Wiley Periodicals, Inc.

Deep learning for automatic Gleason pattern classification for grade group determination of prostate biopsies.

Histopathologic grading of prostate cancer using Gleason patterns (GPs) is subject to a large inter-observer variability, which may result in suboptimal treatment of patients. With the introduction of digitization and whole-slide images of prostate biopsies, computer-aided grading becomes feasible. Computer-aided grading has the potential to improve histopathological grading and treatment selection for prostate cancer. Automated detection of GPs and determination of the grade groups (GG) using a convolutional neural network. In total, 96 prostate biopsies from 38 patients are annotated on pixel-level. Automated detection of GP 3 and GP ≥ 4 in digitized prostate biopsies is performed by re-training the Inception-v3 convolutional neural network (CNN). The outcome of the CNN is subsequently converted into probability maps of GP ≥ 3 and GP ≥ 4, and the GG of the whole biopsy is obtained according to these probability maps. Differentiation between non-atypical and malignant (GP ≥ 3) areas resulted in an accuracy of 92% with a sensitivity and specificity of 90 and 93%, respectively. The differentiation between GP ≥ 4 and GP ≤ 3 was accurate for 90%, with a sensitivity and specificity of 77 and 94%, respectively. Concordance of our automated GG determination method with a genitourinary pathologist was obtained in 65% (κ = 0.70), indicating substantial agreement. A CNN allows for accurate differentiation between non-atypical and malignant areas as defined by GPs, leading to a substantial agreement with the pathologist in defining the GG.

Three-dimensional histopathological reconstruction of bladder tumours.

BACKGROUND: Histopathological analysis is the cornerstone in bladder cancer (BCa) diagnosis. These analysis suffer from a moderate observer agreement in the staging of bladder cancer. Three-dimensional reconstructions have the potential to support the pathologists in visualizing spatial arrangements of structures, which may improve the interpretation of specimen. The aim of this study is to present three-dimensional (3D) reconstructions of histology images. METHODS: En-bloc specimens of transurethral bladder tumour resections were formalin fixed and paraffin embedded. Specimens were cut into sections of 4 µm and stained with Hematoxylin and Eosin (H&E). With a Phillips IntelliSite UltraFast scanner, glass slides were digitized at 20x magnification. The digital images were aligned by performing rigid and affine image alignment. The tumour and the muscularis propria (MP) were manually delineated to create 3D segmentations. In conjunction with a 3D display, the results were visualized with the Vesalius3D interactive visualization application for a 3D workstation. RESULTS: En-bloc resection was performed in 21 BCa patients. Per case, 26-30 sections were included for the reconstruction into a 3D volume. Five cases were excluded due to export problems, size of the dataset or condition of the tissue block. Qualitative evaluation suggested an accurate registration for 13 out of 16 cases. The segmentations allowed full 3D visualization and evaluation of the spatial relationship of the BCa tumour and the MP. CONCLUSION: Digital scanning of en-bloc resected specimens allows a full-fledged 3D reconstruction and analysis and has a potential role to support pathologists in the staging of BCa.

Histopathology: ditch the slides, because digital and 3D are on show.

Due to the growing field of digital pathology, more and more digital histology slides are becoming available. This improves the accessibility, allows teleconsultations from specialized pathologists, improves education, and might give urologist the possibility to review the slides in patient management systems. Moreover, by stacking multiple two-dimensional (2D) digital slides, three-dimensional volumes can be created, allowing improved insight in the growth pattern of a tumor. With the addition of computer-aided diagnosis systems, pathologist can be guided to regions of interest, potentially reducing the workload and interobserver variation. Digital (3D) pathology has the potential to improve dialog between the pathologist and urologist, and, therefore, results in a better treatment selection for urologic patients.

Characteristics of Misclassified CT Perfusion Ischemic Core in Patients with Acute Ischemic Stroke.

BACKGROUND: CT perfusion (CTP) is used to estimate the extent of ischemic core and penumbra in patients with acute ischemic stroke. CTP reliability, however, is limited. This study aims to identify regions misclassified as ischemic core on CTP, using infarct on follow-up noncontrast CT. We aim to assess differences in volumetric and perfusion characteristics in these regions compared to areas that ended up as infarct on follow-up. MATERIALS AND METHODS: This study included 35 patients with >100 mm brain coverage CTP. CTP processing was performed using Philips software (IntelliSpace 7.0). Final infarct was automatically segmented on follow-up noncontrast CT and used as reference. CTP and follow-up noncontrast CT image data were registered. This allowed classification of ischemic lesion agreement (core on CTP: rMTT≥145%, aCBV<2.0 ml/100g and infarct on follow-up noncontrast CT) and misclassified ischemic core (core on CTP, not identified on follow-up noncontrast CT) regions. False discovery ratio (FDR), defined as misclassified ischemic core volume divided by total CTP ischemic core volume, was calculated. Absolute and relative CTP parameters (CBV, CBF, and MTT) were calculated for both misclassified CTP ischemic core and ischemic lesion agreement regions and compared using paired rank-sum tests. RESULTS: Median total CTP ischemic core volume was 49.7ml (IQR:29.9ml-132ml); median misclassified ischemic core volume was 30.4ml (IQR:20.9ml-77.0ml). Median FDR between patients was 62% (IQR:49%-80%). Median relative mean transit time was 243% (IQR:198%-289%) and 342% (IQR:249%-432%) for misclassified and ischemic lesion agreement regions, respectively. Median absolute cerebral blood volume was 1.59 (IQR:1.43-1.79) ml/100g (P<0.01) and 1.38 (IQR:1.15-1.49) ml/100g (P<0.01) for misclassified ischemic core and ischemic lesion agreement, respectively. All CTP parameter values differed significantly. CONCLUSION: For all patients a considerable region of the CTP ischemic core is misclassified. CTP parameters significantly differed between ischemic lesion agreement and misclassified CTP ischemic core, suggesting that CTP analysis may benefit from revisions.