Gigapixel end-to-end training using streaming and attention.

Current hardware limitations make it impossible to train convolutional neural networks on gigapixel image inputs directly. Recent developments in weakly supervised learning, such as attention-gated multiple instance learning, have shown promising results, but often use multi-stage or patch-wise training strategies risking suboptimal feature extraction, which can negatively impact performance. In this paper, we propose to train a ResNet-34 encoder with an attention-gated classification head in an end-to-end fashion, which we call StreamingCLAM, using a streaming implementation of convolutional layers. This allows us to train end-to-end on 4-gigapixel microscopic images using only slide-level labels. We achieve a mean area under the receiver operating characteristic curve of 0.9757 for metastatic breast cancer detection (CAMELYON16), close to fully supervised approaches using pixel-level annotations. Our model can also detect MYC-gene translocation in histologic slides of diffuse large B-cell lymphoma, achieving a mean area under the ROC curve of 0.8259. Furthermore, we show that our model offers a degree of interpretability through the attention mechanism.

Artificial Intelligence Predictive Model for Hormone Therapy Use in Prostate Cancer.

Background: Androgen deprivation therapy (ADT) with radiotherapy can benefit patients with localized prostate cancer. However, ADT can negatively impact quality of life and there remain no validated predictive models to guide its use. Methods: Digital pathology image and clinical data from pre-treatment prostate tissue from 5,727 patients enrolled on five phase III randomized trials treated with radiotherapy +/- ADT were used to develop and validate an artificial intelligence (AI)-derived predictive model to assess ADT benefit with the primary endpoint of distant metastasis. After the model was locked, validation was performed on NRG/RTOG 9408 (n = 1,594) that randomized men to radiotherapy +/- 4 months of ADT. Fine-Gray regression and restricted mean survival times were used to assess the interaction between treatment and predictive model and within predictive model positive and negative subgroup treatment effects. Results: In the NRG/RTOG 9408 validation cohort (14.9 years of median follow-up), ADT significantly improved time to distant metastasis (subdistribution hazard ratio [sHR] = 0.64, 95%CI [0.45-0.90], p = 0.01). The predictive model-treatment interaction was significant (p-interaction = 0.01). In predictive model positive patients (n = 543, 34%), ADT significantly reduced the risk of distant metastasis compared to radiotherapy alone (sHR = 0.34, 95%CI [0.19-0.63], p < 0.001). There were no significant differences between treatment arms in the predictive model negative subgroup (n = 1,051, 66%; sHR = 0.92, 95%CI [0.59-1.43], p = 0.71). Conclusions: Our data, derived and validated from completed randomized phase III trials, show that an AI-based predictive model was able to identify prostate cancer patients, with predominately intermediate-risk disease, who are likely to benefit from short-term ADT.

Artificial Intelligence Predictive Model for Hormone Therapy Use in Prostate Cancer.

Predictive Model for Hormone Therapy in Prostate CancerDigital pathology images and clinical data from pretreatment prostate tissue were used to generate a predictive model to determine patients who would benefit from androgen deprivation therapy (ADT). In model-positive patients, ADT significantly reduced the risk of distant metastasis compared with radiotherapy alone.

Predicting biochemical recurrence of prostate cancer with artificial intelligence.

Background: The first sign of metastatic prostate cancer after radical prostatectomy is rising PSA levels in the blood, termed biochemical recurrence. The prediction of recurrence relies mainly on the morphological assessment of prostate cancer using the Gleason grading system. However, in this system, within-grade morphological patterns and subtle histopathological features are currently omitted, leaving a significant amount of prognostic potential unexplored. Methods: To discover additional prognostic information using artificial intelligence, we trained a deep learning system to predict biochemical recurrence from tissue in H&E-stained microarray cores directly. We developed a morphological biomarker using convolutional neural networks leveraging a nested case-control study of 685 patients and validated on an independent cohort of 204 patients. We use concept-based explainability methods to interpret the learned tissue patterns. Results: The biomarker provides a strong correlation with biochemical recurrence in two sets (n = 182 and n = 204) from separate institutions. Concept-based explanations provided tissue patterns interpretable by pathologists. Conclusions: These results show that the model finds predictive power in the tissue beyond the morphological ISUP grading.

Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge.

Artificial intelligence (AI) has shown promise for diagnosing prostate cancer in biopsies. However, results have been limited to individual studies, lacking validation in multinational settings. Competitions have been shown to be accelerators for medical imaging innovations, but their impact is hindered by lack of reproducibility and independent validation. With this in mind, we organized the PANDA challenge-the largest histopathology competition to date, joined by 1,290 developers-to catalyze development of reproducible AI algorithms for Gleason grading using 10,616 digitized prostate biopsies. We validated that a diverse set of submitted algorithms reached pathologist-level performance on independent cross-continental cohorts, fully blinded to the algorithm developers. On United States and European external validation sets, the algorithms achieved agreements of 0.862 (quadratically weighted κ, 95% confidence interval (CI), 0.840-0.884) and 0.868 (95% CI, 0.835-0.900) with expert uropathologists. Successful generalization across different patient populations, laboratories and reference standards, achieved by a variety of algorithmic approaches, warrants evaluating AI-based Gleason grading in prospective clinical trials.

Streaming Convolutional Neural Networks for End-to-End Learning With Multi-Megapixel Images.

Due to memory constraints on current hardware, most convolution neural networks (CNN) are trained on sub-megapixel images. For example, most popular datasets in computer vision contain images much less than a megapixel in size (0.09MP for ImageNet and 0.001MP for CIFAR-10). In some domains such as medical imaging, multi-megapixel images are needed to identify the presence of disease accurately. We propose a novel method to directly train convolutional neural networks using any input image size end-to-end. This method exploits the locality of most operations in modern convolutional neural networks by performing the forward and backward pass on smaller tiles of the image. In this work, we show a proof of concept using images of up to 66-megapixels (8192×8192), saving approximately 50GB of memory per image. Using two public challenge datasets, we demonstrate that CNNs can learn to extract relevant information from these large images and benefit from increasing resolution. We improved the area under the receiver-operating characteristic curve from 0.580 (4MP) to 0.706 (66MP) for metastasis detection in breast cancer (CAMELYON17). We also obtained a Spearman correlation metric approaching state-of-the-art performance on the TUPAC16 dataset, from 0.485 (1MP) to 0.570 (16MP). Code to reproduce a subset of the experiments is available at https://github.com/DIAGNijmegen/StreamingCNN.

Using deep learning for quantification of cellularity and cell lineages in bone marrow biopsies and comparison to normal age-related variation.

Cellularity estimation forms an important aspect of the visual examination of bone marrow biopsies. In clinical practice, cellularity is estimated by eye under a microscope, which is rapid, but subjective and subject to inter- and intraobserver variability. In addition, there is little consensus in the literature on the normal variation of cellularity with age. Digital image analysis may be used for more objective quantification of cellularity. As such, we developed a deep neural network for the segmentation of six major cell and tissue types in digitized bone marrow trephine biopsies. Using this segmentation, we calculated the overall bone marrow cellularity in a series of biopsies from 130 patients across a wide age range. Using intraclass correlation coefficients (ICC), we measured the agreement between the quantification by the neural network and visual estimation by two pathologists and compared it to baseline human performance. We also examined the age-related changes of cellularity and cell lineages in bone marrow and compared our results to those found in the literature. The network was capable of accurate segmentation (average accuracy and dice score of 0.95 and 0.76, respectively). There was good neural network-pathologist agreement on cellularity measurements (ICC=0.78, 95% CI 0.58-0.85). We found a statistically significant downward trend for cellularity, myelopoiesis and megakaryocytes with age in our cohort. The mean cellularity began at approximately 50% in the third decade of life and then decreased ±2% per decade to 40% in the seventh and eighth decade, but the normal range was very wide (30-70%).

Artificial Intelligence for Diagnosis and Gleason Grading of Prostate Cancer in Biopsies-Current Status and Next Steps.

Diagnosis and Gleason grading of prostate cancer in biopsies are critical for the clinical management of men with prostate cancer. Despite this, the high grading variability among pathologists leads to the potential for under- and overtreatment. Artificial intelligence (AI) systems have shown promise in assisting pathologists to perform Gleason grading, which could help address this problem. In this mini-review, we highlight studies reporting on the development of AI systems for cancer detection and Gleason grading, and discuss the progress needed for widespread clinical implementation, as well as anticipated future developments. PATIENT SUMMARY: This mini-review summarizes the evidence relating to the validation of artificial intelligence (AI)-assisted cancer detection and Gleason grading of prostate cancer in biopsies, and highlights the remaining steps required prior to its widespread clinical implementation. We found that, although there is strong evidence to show that AI is able to perform Gleason grading on par with experienced uropathologists, more work is needed to ensure the accuracy of results from AI systems in diverse settings across different patient populations, digitization platforms, and pathology laboratories.

Detection of Prostate Cancer in Whole-Slide Images Through End-to-End Training With Image-Level Labels.

Prostate cancer is the most prevalent cancer among men in Western countries, with 1.1 million new diagnoses every year. The gold standard for the diagnosis of prostate cancer is a pathologists' evaluation of prostate tissue. To potentially assist pathologists deep / learning / based cancer detection systems have been developed. Many of the state-of-the-art models are patch / based convolutional neural networks, as the use of entire scanned slides is hampered by memory limitations on accelerator cards. Patch-based systems typically require detailed, pixel-level annotations for effective training. However, such annotations are seldom readily available, in contrast to the clinical reports of pathologists, which contain slide-level labels. As such, developing algorithms which do not require manual pixel-wise annotations, but can learn using only the clinical report would be a significant advancement for the field. In this paper, we propose to use a streaming implementation of convolutional layers, to train a modern CNN (ResNet / 34) with 21 million parameters end-to-end on 4712 prostate biopsies. The method enables the use of entire biopsy images at high-resolution directly by reducing the GPU memory requirements by 2.4 TB. We show that modern CNNs, trained using our streaming approach, can extract meaningful features from high-resolution images without additional heuristics, reaching similar performance as state-of-the-art patch-based and multiple-instance learning methods. By circumventing the need for manual annotations, this approach can function as a blueprint for other tasks in histopathological diagnosis. The source code to reproduce the streaming models is available at https://github.com/DIAGNijmegen/ pathology-streaming-pipeline.

Automated deep-learning system for Gleason grading of prostate cancer using biopsies: a diagnostic study.

BACKGROUND: The Gleason score is the strongest correlating predictor of recurrence for prostate cancer, but has substantial inter-observer variability, limiting its usefulness for individual patients. Specialised urological pathologists have greater concordance; however, such expertise is not widely available. Prostate cancer diagnostics could thus benefit from robust, reproducible Gleason grading. We aimed to investigate the potential of deep learning to perform automated Gleason grading of prostate biopsies. METHODS: In this retrospective study, we developed a deep-learning system to grade prostate biopsies following the Gleason grading standard. The system was developed using randomly selected biopsies, sampled by the biopsy Gleason score, from patients at the Radboud University Medical Center (pathology report dated between Jan 1, 2012, and Dec 31, 2017). A semi-automatic labelling technique was used to circumvent the need for manual annotations by pathologists, using pathologists' reports as the reference standard during training. The system was developed to delineate individual glands, assign Gleason growth patterns, and determine the biopsy-level grade. For validation of the method, a consensus reference standard was set by three expert urological pathologists on an independent test set of 550 biopsies. Of these 550, 100 were used in an observer experiment, in which the system, 13 pathologists, and two pathologists in training were compared with respect to the reference standard. The system was also compared to an external test dataset of 886 cores, which contained 245 cores from a different centre that were independently graded by two pathologists. FINDINGS: We collected 5759 biopsies from 1243 patients. The developed system achieved a high agreement with the reference standard (quadratic Cohen's kappa 0·918, 95% CI 0·891-0·941) and scored highly at clinical decision thresholds: benign versus malignant (area under the curve 0·990, 95% CI 0·982-0·996), grade group of 2 or more (0·978, 0·966-0·988), and grade group of 3 or more (0·974, 0·962-0·984). In an observer experiment, the deep-learning system scored higher (kappa 0·854) than the panel (median kappa 0·819), outperforming 10 of 15 pathologist observers. On the external test dataset, the system obtained a high agreement with the reference standard set independently by two pathologists (quadratic Cohen's kappa 0·723 and 0·707) and within inter-observer variability (kappa 0·71). INTERPRETATION: Our automated deep-learning system achieved a performance similar to pathologists for Gleason grading and could potentially contribute to prostate cancer diagnosis. The system could potentially assist pathologists by screening biopsies, providing second opinions on grade group, and presenting quantitative measurements of volume percentages. FUNDING: Dutch Cancer Society.

Learning to detect lymphocytes in immunohistochemistry with deep learning.

The immune system is of critical importance in the development of cancer. The evasion of destruction by the immune system is one of the emerging hallmarks of cancer. We have built a dataset of 171,166 manually annotated CD3+ and CD8+ cells, which we used to train deep learning algorithms for automatic detection of lymphocytes in histopathology images to better quantify immune response. Moreover, we investigate the effectiveness of four deep learning based methods when different subcompartments of the whole-slide image are considered: normal tissue areas, areas with immune cell clusters, and areas containing artifacts. We have compared the proposed methods in breast, colon and prostate cancer tissue slides collected from nine different medical centers. Finally, we report the results of an observer study on lymphocyte quantification, which involved four pathologists from different medical centers, and compare their performance with the automatic detection. The results give insights on the applicability of the proposed methods for clinical use. U-Net obtained the highest performance with an F1-score of 0.78 and the highest agreement with manual evaluation (κ=0.72), whereas the average pathologists agreement with reference standard was κ=0.64. The test set and the automatic evaluation procedure are publicly available at lyon19.grand-challenge.org.

Deep learning assisted mitotic counting for breast cancer.

As part of routine histological grading, for every invasive breast cancer the mitotic count is assessed by counting mitoses in the (visually selected) region with the highest proliferative activity. Because this procedure is prone to subjectivity, the present study compares visual mitotic counting with deep learning based automated mitotic counting and fully automated hotspot selection. Two cohorts were used in this study. Cohort A comprised 90 prospectively included tumors which were selected based on the mitotic frequency scores given during routine glass slide diagnostics. This pathologist additionally assessed the mitotic count in these tumors in whole slide images (WSI) within a preselected hotspot. A second observer performed the same procedures on this cohort. The preselected hotspot was generated by a convolutional neural network (CNN) trained to detect all mitotic figures in digitized hematoxylin and eosin (H&E) sections. The second cohort comprised a multicenter, retrospective TNBC cohort (n = 298), of which the mitotic count was assessed by three independent observers on glass slides. The same CNN was applied on this cohort and the absolute number of mitotic figures in the hotspot was compared to the averaged mitotic count of the observers. Baseline interobserver agreement for glass slide assessment in cohort A was good (kappa 0.689; 95% CI 0.580-0.799). Using the CNN generated hotspot in WSI, the agreement score increased to 0.814 (95% CI 0.719-0.909). Automated counting by the CNN in comparison with observers counting in the predefined hotspot region yielded an average kappa of 0.724. We conclude that manual mitotic counting is not affected by assessment modality (glass slides, WSI) and that counting mitotic figures in WSI is feasible. Using a predefined hotspot area considerably improves reproducibility. Also, fully automated assessment of mitotic score appears to be feasible without introducing additional bias or variability.