A new optical density granulometry-based descriptor for the classification of prostate histological images using shallow and deep Gaussian processes

The CrowdGleason dataset: Learning the Gleason grade from crowds and experts

BACKGROUND

Currently, prostate cancer (PCa) diagnosis relies on the human analysis of prostate biopsy Whole Slide Images (WSIs) using the Gleason score. Since this process is error-prone and time-consuming, recent advances in machine learning have promoted the use of automated systems to assist pathologists. Unfortunately, labeled datasets for training and validation are scarce due to the need for expert pathologists to provide ground-truth labels.

METHODS

This work introduces a new prostate histopathological dataset named CrowdGleason, which consists of 19,077 patches from 1045 WSIs with various Gleason grades. The dataset was annotated using a crowdsourcing protocol involving seven pathologists-in-training to distribute the labeling effort. To provide a baseline analysis, two crowdsourcing methods based on Gaussian Processes (GPs) were evaluated for Gleason grade prediction: SVGPCR, which learns a model from the CrowdGleason dataset, and SVGPMIX, which combines data from the public dataset SICAPv2 and the CrowdGleason dataset. The performance of these methods was compared with other crowdsourcing and expert label-based methods through comprehensive experiments.

RESULTS

The results demonstrate that our GP-based crowdsourcing approach outperforms other methods for aggregating crowdsourced labels (κ=0.7048±0.0207) for SVGPCR vs.(κ=0.6576±0.0086) for SVGP with majority voting). SVGPCR trained with crowdsourced labels performs better than GP trained with expert labels from SICAPv2 (κ=0.6583±0.0220) and outperforms most individual pathologists-in-training (mean κ=0.5432). Additionally, SVGPMIX trained with a combination of SICAPv2 and CrowdGleason achieves the highest performance on both datasets (κ=0.7814±0.0083 and κ=0.7276±0.0260).

CONCLUSION

The experiments show that the CrowdGleason dataset can be successfully used for training and validating supervised and crowdsourcing methods. Furthermore, the crowdsourcing methods trained on this dataset obtain competitive results against those using expert labels. Interestingly, the combination of expert and non-expert labels opens the door to a future of massive labeling by incorporating both expert and non-expert pathologist annotators.

All you need is data preparation: A systematic review of image harmonization techniques in Multi-center/device studies for medical support systems

BACKGROUND AND OBJECTIVES

Artificial intelligence (AI) models trained on multi-centric and multi-device studies can provide more robust insights and research findings compared to single-center studies. However, variability in acquisition protocols and equipment can introduce inconsistencies that hamper the effective pooling of multi-source datasets. This systematic review evaluates strategies for image harmonization, which standardizes appearances to enable reliable AI analysis of multi-source medical imaging.

METHODS

A literature search using PRISMA guidelines was conducted to identify relevant papers published between 2013 and 2023 analyzing multi-centric and multi-device medical imaging studies that utilized image harmonization approaches.

RESULTS

Common image harmonization techniques included grayscale normalization (improving classification accuracy by up to 24.42 %), resampling (increasing the percentage of robust radiomics features from 59.5 % to 89.25 %), and color normalization (enhancing AUC by up to 0.25 in external test sets). Initially, mathematical and statistical methods dominated, but machine and deep learning adoption has risen recently. Color imaging modalities like digital pathology and dermatology have remained prominent application areas, though harmonization efforts have expanded to diverse fields including radiology, nuclear medicine, and ultrasound imaging. In all the modalities covered by this review, image harmonization improved AI performance, with increasing of up to 24.42 % in classification accuracy and 47 % in segmentation Dice scores.

CONCLUSIONS

Continued progress in image harmonization represents a promising strategy for advancing healthcare by enabling large-scale, reliable analysis of integrated multi-source datasets using AI. Standardizing imaging data across clinical settings can help realize personalized, evidence-based care supported by data-driven technologies while mitigating biases associated with specific populations or acquisition protocols.

Artificial intelligence in digital pathology: a systematic review and meta-analysis of diagnostic test accuracy

Ensuring diagnostic performance of artificial intelligence (AI) before introduction into clinical practice is essential. Growing numbers of studies using AI for digital pathology have been reported over recent years. The aim of this work is to examine the diagnostic accuracy of AI in digital pathology images for any disease. This systematic review and meta-analysis included diagnostic accuracy studies using any type of AI applied to whole slide images (WSIs) for any disease. The reference standard was diagnosis by histopathological assessment and/or immunohistochemistry. Searches were conducted in PubMed, EMBASE and CENTRAL in June 2022. Risk of bias and concerns of applicability were assessed using the QUADAS-2 tool. Data extraction was conducted by two investigators and meta-analysis was performed using a bivariate random effects model, with additional subgroup analyses also performed. Of 2976 identified studies, 100 were included in the review and 48 in the meta-analysis. Studies were from a range of countries, including over 152,000 whole slide images (WSIs), representing many diseases. These studies reported a mean sensitivity of 96.3% (CI 94.1-97.7) and mean specificity of 93.3% (CI 90.5-95.4). There was heterogeneity in study design and 99% of studies identified for inclusion had at least one area at high or unclear risk of bias or applicability concerns. Details on selection of cases, division of model development and validation data and raw performance data were frequently ambiguous or missing. AI is reported as having high diagnostic accuracy in the reported areas but requires more rigorous evaluation of its performance.

Are you sure it's an artifact? Artifact detection and uncertainty quantification in histological images

Modern cancer diagnostics involves extracting tissue specimens from suspicious areas and conducting histotechnical procedures to prepare a digitized glass slide, called Whole Slide Image (WSI), for further examination. These procedures frequently introduce different types of artifacts in the obtained WSI, and histological artifacts might influence Computational Pathology (CPATH) systems further down to a diagnostic pipeline if not excluded or handled. Deep Convolutional Neural Networks (DCNNs) have achieved promising results for the detection of some WSI artifacts, however, they do not incorporate uncertainty in their predictions. This paper proposes an uncertainty-aware Deep Kernel Learning (DKL) model to detect blurry areas and folded tissues, two types of artifacts that can appear in WSIs. The proposed probabilistic model combines a CNN feature extractor and a sparse Gaussian Processes (GPs) classifier, which improves the performance of current state-of-the-art artifact detection DCNNs and provides uncertainty estimates. We achieved 0.996 and 0.938 F1 scores for blur and folded tissue detection on unseen data, respectively. In extensive experiments, we validated the DKL model on unseen data from external independent cohorts with different staining and tissue types, where it outperformed DCNNs. Interestingly, the DKL model is more confident in the correct predictions and less in the wrong ones. The proposed DKL model can be integrated into the preprocessing pipeline of CPATH systems to provide reliable predictions and possibly serve as a quality control tool.

Label-free multimodal electro-thermo-mechanical (ETM) phenotyping as a novel biomarker to differentiate between normal, benign, and cancerous breast biopsy tissues

BACKGROUND

Technologies for quick and label-free diagnosis of malignancies from breast tissues have the potential to be a significant adjunct to routine diagnostics. The biophysical phenotypes of breast tissues, such as its electrical, thermal, and mechanical properties (ETM), have the potential to serve as novel markers to differentiate between normal, benign, and malignant tissue.

RESULTS

We report a system-of-biochips (SoB) integrated into a semi-automated mechatronic system that can characterize breast biopsy tissues using electro-thermo-mechanical sensing. The SoB, fabricated on silicon using microfabrication techniques, can measure the electrical impedance (Z), thermal conductivity (K), mechanical stiffness (k), and viscoelastic stress relaxation (%R) of the samples. The key sensing elements of the biochips include interdigitated electrodes, resistance temperature detectors, microheaters, and a micromachined diaphragm with piezoresistive bridges. Multi-modal ETM measurements performed on formalin-fixed tumour and adjacent normal breast biopsy samples from N = 14 subjects were able to differentiate between invasive ductal carcinoma (malignant), fibroadenoma (benign), and adjacent normal (healthy) tissues with a root mean square error of 0.2419 using a Gaussian process classifier. Carcinoma tissues were observed to have the highest mean impedance (110018.8 ± 20293.8 Ω) and stiffness (0.076 ± 0.009 kNm-1) and the lowest thermal conductivity (0.189 ± 0.019 Wm-1 K-1) amongst the three groups, while the fibroadenoma samples had the highest percentage relaxation in normalized load (47.8 ± 5.12%).

CONCLUSIONS

The work presents a novel strategy to characterize the multi-modal biophysical phenotype of breast biopsy tissues to aid in cancer diagnosis from small-sized tumour samples. The methodology envisions to supplement the existing technology gap in the analysis of breast tissue samples in the pathology laboratories to aid the diagnostic workflow.

An update on computational pathology tools for genitourinary pathology practice: A review paper from the Genitourinary Pathology Society (GUPS)

Machine learning has been leveraged for image analysis applications throughout a multitude of subspecialties. This position paper provides a perspective on the evolutionary trajectory of practical deep learning tools for genitourinary pathology through evaluating the most recent iterations of such algorithmic devices. Deep learning tools for genitourinary pathology demonstrate potential to enhance prognostic and predictive capacity for tumor assessment including grading, staging, and subtype identification, yet limitations in data availability, regulation, and standardization have stymied their implementation.

Applications of Digital and Computational Pathology and Artificial Intelligence in Genitourinary Pathology Diagnostics

As machine learning (ML) solutions for genitourinary pathology image analysis are fostered by a progressively digitized laboratory landscape, these integrable modalities usher in a revolution in histopathological diagnosis. As technology advances, limitations stymying clinical artificial intelligence (AI) will not be extinguished without thorough validation and interrogation of ML tools by pathologists and regulatory bodies alike. ML solutions deployed in clinical settings for applications in prostate pathology yield promising results. Recent breakthroughs in clinical artificial intelligence for genitourinary pathology demonstrate unprecedented generalizability, heralding prospects for a future in which AI-driven assistive solutions may be seen as laboratory faculty, rather than novelty.

Prediction of Adverse Pathology at Radical Prostatectomy in Grade Group 2 and 3 Prostate Biopsies Using Machine Learning

PURPOSE

There is ongoing clinical need to improve estimates of disease outcome in prostate cancer. Machine learning (ML) approaches to pathologic diagnosis and prognosis are a promising and increasingly used strategy. In this study, we use an ML algorithm for prediction of adverse outcomes at radical prostatectomy (RP) using whole-slide images (WSIs) of prostate biopsies with Grade Group (GG) 2 or 3 disease.

METHODS

We performed a retrospective review of prostate biopsies collected at our institution which had corresponding RP, GG 2 or 3 disease one or more cores, and no biopsies with higher than GG 3 disease. A hematoxylin and eosin-stained core needle biopsy from each site with GG 2 or 3 disease was scanned and used as the sole input for the algorithm. The ML pipeline had three phases: image preprocessing, feature extraction, and adverse outcome prediction. First, patches were extracted from each biopsy scan. Subsequently, the pre-trained Visual Geometry Group-16 convolutional neural network was used for feature extraction. A representative feature vector was then used as input to an Extreme Gradient Boosting classifier for predicting the binary adverse outcome. We subsequently assessed patient clinical risk using CAPRA score for comparison with the ML pipeline results.

RESULTS

The data set included 361 WSIs from 107 patients (56 with adverse pathology at RP). The area under the receiver operating characteristic curves for the ML classification were 0.72 (95% CI, 0.62 to 0.81), 0.65 (95% CI, 0.53 to 0.79) and 0.89 (95% CI, 0.79 to 1.00) for the entire cohort, and GG 2 and GG 3 patients, respectively, similar to the performance of the CAPRA clinical risk assessment.

CONCLUSION

We provide evidence for the potential of ML algorithms to use WSIs of needle core prostate biopsies to estimate clinically relevant prostate cancer outcomes.

Cultivating Clinical Clarity through Computer Vision: A Current Perspective on Whole Slide Imaging and Artificial Intelligence

Diagnostic devices, methodological approaches, and traditional constructs of clinical pathology practice, cultivated throughout centuries, have transformed radically in the wake of explosive technological growth and other, e.g., environmental, catalysts of change. Ushered into the fray of modern laboratory medicine are digital imaging devices and machine-learning (ML) software fashioned to mitigate challenges, e.g., practitioner shortage while preparing clinicians for emerging interconnectivity of environments and diagnostic information in the era of big data. As computer vision shapes new constructs for the modern world and intertwines with clinical medicine, cultivating clarity of our new terrain through examining the trajectory and current scope of computational pathology and its pertinence to clinical practice is vital. Through review of numerous studies, we find developmental efforts for ML migrating from research to standardized clinical frameworks while overcoming obstacles that have formerly curtailed adoption of these tools, e.g., generalizability, data availability, and user-friendly accessibility. Groundbreaking validatory efforts have facilitated the clinical deployment of ML tools demonstrating the capacity to effectively aid in distinguishing tumor subtype and grade, classify early vs. advanced cancer stages, and assist in quality control and primary diagnosis applications. Case studies have demonstrated the benefits of streamlined, digitized workflows for practitioners alleviated by decreased burdens.

Proportion constrained weakly supervised histopathology image classification

Multiple instance learning (MIL) deals with data grouped into bags of instances, of which only the global information is known. In recent years, this weakly supervised learning paradigm has become very popular in histological image analysis because it alleviates the burden of labeling all cancerous regions of large Whole Slide Images (WSIs) in detail. However, these methods require large datasets to perform properly, and many approaches only focus on simple binary classification. This often does not match the real-world problems where multi-label settings are frequent and possible constraints must be taken into account. In this work, we propose a novel multi-label MIL formulation based on inequality constraints that is able to incorporate prior knowledge about instance proportions. Our method has a theoretical foundation in optimization with log-barrier extensions, applied to bag-level class proportions. This encourages the model to respect the proportion ordering during training. Extensive experiments on a new public dataset of prostate cancer WSIs analysis, SICAP-MIL, demonstrate that using the prior proportion information we can achieve instance-level results similar to supervised methods on datasets of similar size. In comparison with prior MIL settings, our method allows for ∼13% improvements in instance-level accuracy, and ∼3% in the multi-label mean area under the ROC curve at the bag-level.

Deep Gaussian processes for multiple instance learning: Application to CT intracranial hemorrhage detection

BACKGROUND AND OBJECTIVE

Intracranial hemorrhage (ICH) is a life-threatening emergency that can lead to brain damage or death, with high rates of mortality and morbidity. The fast and accurate detection of ICH is important for the patient to get an early and efficient treatment. To improve this diagnostic process, the application of Deep Learning (DL) models on head CT scans is an active area of research. Although promising results have been obtained, many of the proposed models require slice-level annotations by radiologists, which are costly and time-consuming.

METHODS

We formulate the ICH detection as a problem of Multiple Instance Learning (MIL) that allows training with only scan-level annotations. We develop a new probabilistic method based on Deep Gaussian Processes (DGP) that is able to train with this MIL setting and accurately predict ICH at both slice- and scan-level. The proposed DGPMIL model is able to capture complex feature relations by using multiple Gaussian Process (GP) layers, as we show experimentally.

RESULTS

To highlight the advantages of DGPMIL in a general MIL setting, we first conduct several controlled experiments on the MNIST dataset. We show that multiple GP layers outperform one-layer GP models, especially for complex feature distributions. For ICH detection experiments, we use two public brain CT datasets (RSNA and CQ500). We first train a Convolutional Neural Network (CNN) with an attention mechanism to extract the image features, which are fed into our DGPMIL model to perform the final predictions. The results show that DGPMIL model outperforms VGPMIL as well as the attention-based CNN for MIL and other state-of-the-art methods for this problem. The best performing DGPMIL model reaches an AUC-ROC of 0.957 (resp. 0.909) and an AUC-PR of 0.961 (resp. 0.889) on the RSNA (resp. CQ500) dataset.

CONCLUSION

The competitive performance at slice- and scan-level shows that DGPMIL model provides an accurate diagnosis on slices without the need for slice-level annotations by radiologists during training. As MIL is a common problem setting, our model can be applied to a broader range of other tasks, especially in medical image classification, where it can help the diagnostic process.

Blind color deconvolution, normalization, and classification of histological images using general super Gaussian priors and Bayesian inference

BACKGROUND AND OBJECTIVE

Color variations in digital histopathology severely impact the performance of computer-aided diagnosis systems. They are due to differences in the staining process and acquisition system, among other reasons. Blind color deconvolution techniques separate multi-stained images into single stained bands which, once normalized, can be used to eliminate these negative color variations and improve the performance of machine learning tasks.

METHODS

In this work, we decompose the observed RGB image in its hematoxylin and eosin components. We apply Bayesian modeling and inference based on the use of Super Gaussian sparse priors for each stain together with prior closeness to a given reference color-vector matrix. The hematoxylin and eosin components are then used for image normalization and classification of histological images. The proposed framework is tested on stain separation, image normalization, and cancer classification problems. The results are measured using the peak signal to noise ratio, normalized median intensity and the area under ROC curve on five different databases.

RESULTS

The obtained results show the superiority of our approach to current state-of-the-art blind color deconvolution techniques. In particular, the fidelity to the tissue improves 1,27 dB in mean PSNR. The normalized median intensity shows a good normalization quality of the proposed approach on the tested datasets. Finally, in cancer classification experiments the area under the ROC curve improves from 0.9491 to 0.9656 and from 0.9279 to 0.9541 on Camelyon-16 and Camelyon-17, respectively, when the original and processed images are used. Furthermore, these figures of merits are better than those obtained by the methods compared with.

CONCLUSIONS

The proposed framework for blind color deconvolution, normalization and classification of images guarantees fidelity to the tissue structure and can be used both for normalization and classification. In addition, color deconvolution enables the use of the optical density space for classification, which improves the classification performance.

Self-Learning for Weakly Supervised Gleason Grading of Local Patterns

Prostate cancer is one of the main diseases affecting men worldwide. The gold standard for diagnosis and prognosis is the Gleason grading system. In this process, pathologists manually analyze prostate histology slides under microscope, in a high time-consuming and subjective task. In the last years, computer-aided-diagnosis (CAD) systems have emerged as a promising tool that could support pathologists in the daily clinical practice. Nevertheless, these systems are usually trained using tedious and prone-to-error pixel-level annotations of Gleason grades in the tissue. To alleviate the need of manual pixel-wise labeling, just a handful of works have been presented in the literature. Furthermore, despite the promising results achieved on global scoring the location of cancerous patterns in the tissue is only qualitatively addressed. These heatmaps of tumor regions, however, are crucial to the reliability of CAD systems as they provide explainability to the system's output and give confidence to pathologists that the model is focusing on medical relevant features. Motivated by this, we propose a novel weakly-supervised deep-learning model, based on self-learning CNNs, that leverages only the global Gleason score of gigapixel whole slide images during training to accurately perform both, grading of patch-level patterns and biopsy-level scoring. To evaluate the performance of the proposed method, we perform extensive experiments on three different external datasets for the patch-level Gleason grading, and on two different test sets for global Grade Group prediction. We empirically demonstrate that our approach outperforms its supervised counterpart on patch-level Gleason grading by a large margin, as well as state-of-the-art methods on global biopsy-level scoring. Particularly, the proposed model brings an average improvement on the Cohen's quadratic kappa ( κ) score of nearly 18% compared to full-supervision for the patch-level Gleason grading task. This suggests that the absence of the annotator's bias in our approach and the capability of using large weakly labeled datasets during training leads to higher performing and more robust models. Furthermore, raw features obtained from the patch-level classifier showed to generalize better than previous approaches in the literature to the subjective global biopsy-level scoring.

Artificial intelligence in prostate histopathology: where are we in 2021?

PURPOSE OF REVIEW

Artificial intelligence has made an entrance into mainstream applications of daily life but the clinical deployment of artificial intelligence-supported histological analysis is still at infancy. Recent years have seen a surge in technological advance regarding the use of artificial intelligence in pathology, in particular in the diagnosis of prostate cancer.

RECENT FINDINGS

We review first impressions of how artificial intelligence impacts the clinical performance of pathologists in the analysis of prostate tissue. Several challenges in the deployment of artificial intelligence remain to be overcome. Finally, we discuss how artificial intelligence can help in generating new knowledge that is interpretable by humans.

SUMMARY

It is evident that artificial intelligence has the potential to outperform most pathologists in detecting prostate cancer, and does not suffer from inherent interobserver variability. Nonetheless, large clinical validation studies that unequivocally prove the benefit of artificial intelligence support in pathology are necessary. Regardless, artificial intelligence may soon automate and standardize many facets of routine work, including qualitative (i.e. Gleason Grading) and quantitative measures (i.e. portion of Gleason Grades and tumor volume). For the near future, a model where pathologists are enhanced by second-review or real-time artificial intelligence systems appears to be the most promising approach.

The 2019 Genitourinary Pathology Society (GUPS) White Paper on Contemporary Grading of Prostate Cancer

CONTEXT.—

Controversies and uncertainty persist in prostate cancer grading.

OBJECTIVE.—

To update grading recommendations.

DATA SOURCES.—

Critical review of the literature along with pathology and clinician surveys.

CONCLUSIONS.—

Percent Gleason pattern 4 (%GP4) is as follows: (1) report %GP4 in needle biopsy with Grade Groups (GrGp) 2 and 3, and in needle biopsy on other parts (jars) of lower grade in cases with at least 1 part showing Gleason score (GS) 4 + 4 = 8; and (2) report %GP4: less than 5% or less than 10% and 10% increments thereafter. Tertiary grade patterns are as follows: (1) replace "tertiary grade pattern" in radical prostatectomy (RP) with "minor tertiary pattern 5 (TP5)," and only use in RP with GrGp 2 or 3 with less than 5% Gleason pattern 5; and (2) minor TP5 is noted along with the GS, with the GrGp based on the GS. Global score and magnetic resonance imaging (MRI)-targeted biopsies are as follows: (1) when multiple undesignated cores are taken from a single MRI-targeted lesion, an overall grade for that lesion is given as if all the involved cores were one long core; and (2) if providing a global score, when different scores are found in the standard and the MRI-targeted biopsy, give a single global score (factoring both the systematic standard and the MRI-targeted positive cores). Grade Groups are as follows: (1) Grade Groups (GrGp) is the terminology adopted by major world organizations; and (2) retain GS 3 + 5 = 8 in GrGp 4. Cribriform carcinoma is as follows: (1) report the presence or absence of cribriform glands in biopsy and RP with Gleason pattern 4 carcinoma. Intraductal carcinoma (IDC-P) is as follows: (1) report IDC-P in biopsy and RP; (2) use criteria based on dense cribriform glands (>50% of the gland is composed of epithelium relative to luminal spaces) and/or solid nests and/or marked pleomorphism/necrosis; (3) it is not necessary to perform basal cell immunostains on biopsy and RP to identify IDC-P if the results would not change the overall (highest) GS/GrGp part per case; (4) do not include IDC-P in determining the final GS/GrGp on biopsy and/or RP; and (5) "atypical intraductal proliferation (AIP)" is preferred for an intraductal proliferation of prostatic secretory cells which shows a greater degree of architectural complexity and/or cytological atypia than typical high-grade prostatic intraepithelial neoplasia, yet falling short of the strict diagnostic threshold for IDC-P. Molecular testing is as follows: (1) Ki67 is not ready for routine clinical use; (2) additional studies of active surveillance cohorts are needed to establish the utility of PTEN in this setting; and (3) dedicated studies of RNA-based assays in active surveillance populations are needed to substantiate the utility of these expensive tests in this setting. Artificial intelligence and novel grading schema are as follows: (1) incorporating reactive stromal grade, percent GP4, minor tertiary GP5, and cribriform/intraductal carcinoma are not ready for adoption in current practice.

A multi-resolution model for histopathology image classification and localization with multiple instance learning

Large numbers of histopathological images have been digitized into high resolution whole slide images, opening opportunities in developing computational image analysis tools to reduce pathologists' workload and potentially improve inter- and intra-observer agreement. Most previous work on whole slide image analysis has focused on classification or segmentation of small pre-selected regions-of-interest, which requires fine-grained annotation and is non-trivial to extend for large-scale whole slide analysis. In this paper, we proposed a multi-resolution multiple instance learning model that leverages saliency maps to detect suspicious regions for fine-grained grade prediction. Instead of relying on expensive region- or pixel-level annotations, our model can be trained end-to-end with only slide-level labels. The model is developed on a large-scale prostate biopsy dataset containing 20,229 slides from 830 patients. The model achieved 92.7% accuracy, 81.8% Cohen's Kappa for benign, low grade (i.e. Grade group 1) and high grade (i.e. Grade group ≥ 2) prediction, an area under the receiver operating characteristic curve (AUROC) of 98.2% and an average precision (AP) of 97.4% for differentiating malignant and benign slides. The model obtained an AUROC of 99.4% and an AP of 99.8% for cancer detection on an external dataset.

Going deeper through the Gleason scoring scale: An automatic end-to-end system for histology prostate grading and cribriform pattern detection

BACKGROUND AND OBJECTIVE

Prostate cancer is one of the most common diseases affecting men worldwide. The Gleason scoring system is the primary diagnostic and prognostic tool for prostate cancer. Furthermore, recent reports indicate that the presence of patterns of the Gleason scale such as the cribriform pattern may also correlate with a worse prognosis compared to other patterns belonging to the Gleason grade 4. Current clinical guidelines have indicated the convenience of highlight its presence during the analysis of biopsies. All these requirements suppose a great workload for the pathologist during the analysis of each sample, which is based on the pathologist's visual analysis of the morphology and organisation of the glands in the tissue, a time-consuming and subjective task. In recent years, with the development of digitisation devices, the use of computer vision techniques for the analysis of biopsies has increased. However, to the best of the authors' knowledge, the development of algorithms to automatically detect individual cribriform patterns belonging to Gleason grade 4 has not yet been studied in the literature. The objective of the work presented in this paper is to develop a deep-learning-based system able to support pathologists in the daily analysis of prostate biopsies. This analysis must include the Gleason grading of local structures, the detection of cribriform patterns, and the Gleason scoring of the whole biopsy.

METHODS

The methodological core of this work is a patch-wise predictive model based on convolutional neural networks able to determine the presence of cancerous patterns based on the Gleason grading system. In particular, we train from scratch a simple self-design architecture with three filters and a top model with global-max pooling. The cribriform pattern is detected by retraining the set of filters of the last convolutional layer in the network. Subsequently, a biopsy-level prediction map is reconstructed by bi-linear interpolation of the patch-level prediction of the Gleason grades. In addition, from the reconstructed prediction map, we compute the percentage of each Gleason grade in the tissue to feed a multi-layer perceptron which provides a biopsy-level score.

RESULTS

In our SICAPv2 database, composed of 182 annotated whole slide images, we obtained a Cohen's quadratic kappa of 0.77 in the test set for the patch-level Gleason grading with the proposed architecture trained from scratch. Our results outperform previous ones reported in the literature. Furthermore, this model reaches the level of fine-tuned state-of-the-art architectures in a patient-based four groups cross validation. In the cribriform pattern detection task, we obtained an area under ROC curve of 0.82. Regarding the biopsy Gleason scoring, we achieved a quadratic Cohen's Kappa of 0.81 in the test subset. Shallow CNN architectures trained from scratch outperform current state-of-the-art methods for Gleason grades classification. Our proposed model is capable of characterising the different Gleason grades in prostate tissue by extracting low-level features through three basic blocks (i.e. convolutional layer + max pooling). The use of global-max pooling to reduce each activation map has shown to be a key factor for reducing complexity in the model and avoiding overfitting. Regarding the Gleason scoring of biopsies, a multi-layer perceptron has shown to better model the decision-making of pathologists than previous simpler models used in the literature.