Deep Learning for Natural Language Processing in Urology: State-of-the-Art Automated Extraction of Detailed Pathologic Prostate Cancer Data From Narratively Written Electronic Health Records

Validation of a Zero-shot Learning Natural Language Processing Tool to Facilitate Data Abstraction for Urologic Research

BACKGROUND

Urologic research often requires data abstraction from unstructured text contained within the electronic health record. A number of natural language processing (NLP) tools have been developed to aid with this time-consuming task; however, the generalizability of these tools is typically limited by the need for task-specific training.

OBJECTIVE

To describe the development and validation of a zero-shot learning NLP tool to facilitate data abstraction from unstructured text for use in downstream urologic research.

DESIGN, SETTING, AND PARTICIPANTS

An NLP tool based on the GPT-3.5 model from OpenAI was developed and compared with three physicians for time to task completion and accuracy for abstracting 14 unique variables from a set of 199 deidentified radical prostatectomy pathology reports. The reports were processed in vectorized and scanned formats to establish the impact of optical character recognition on data abstraction.

INTERVENTION

A zero-shot learning NLP tool for data abstraction.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

The tool was compared with the human abstractors in terms of superiority for data abstraction speed and noninferiority for accuracy.

RESULTS AND LIMITATIONS

The human abstractors required a median (interquartile range) of 93 s (72-122 s) per report for data abstraction, whereas the software required a median of 12 s (10-15 s) for the vectorized reports and 15 s (13-17 s) for the scanned reports (p < 0.001 for all paired comparisons). The accuracies of the three human abstractors were 94.7% (95% confidence interval [CI], 93.8-95.5%), 97.8% (95% CI, 97.2-98.3%), and 96.4% (95% CI, 95.6-97%) for the combined set of 2786 data points. The tool had accuracy of 94.2% (95% CI, 93.3-94.9%) for the vectorized reports and was noninferior to the human abstractors at a margin of -10% (α = 0.025). The tool had slightly lower accuracy of 88.7% (95% CI 87.5-89.9%) for the scanned reports, making it noninferior to two of three human abstractors.

CONCLUSIONS

The developed zero-shot learning NLP tool offers urologic researchers a highly generalizable and accurate method for data abstraction from unstructured text. An open access version of the tool is available for immediate use by the urologic community.

PATIENT SUMMARY

In this report, we describe the design and validation of an artificial intelligence tool for abstracting discrete data from unstructured notes contained within the electronic medical record. This freely available tool, which is based on the GPT-3.5 technology from OpenAI, is intended to facilitate research and scientific discovery by the urologic community.

Extracting structured information from unstructured histopathology reports using generative pre-trained transformer 4 (GPT-4)

Deep learning applied to whole-slide histopathology images (WSIs) has the potential to enhance precision oncology and alleviate the workload of experts. However, developing these models necessitates large amounts of data with ground truth labels, which can be both time-consuming and expensive to obtain. Pathology reports are typically unstructured or poorly structured texts, and efforts to implement structured reporting templates have been unsuccessful, as these efforts lead to perceived extra workload. In this study, we hypothesised that large language models (LLMs), such as the generative pre-trained transformer 4 (GPT-4), can extract structured data from unstructured plain language reports using a zero-shot approach without requiring any re-training. We tested this hypothesis by utilising GPT-4 to extract information from histopathological reports, focusing on two extensive sets of pathology reports for colorectal cancer and glioblastoma. We found a high concordance between LLM-generated structured data and human-generated structured data. Consequently, LLMs could potentially be employed routinely to extract ground truth data for machine learning from unstructured pathology reports in the future. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Expanding the Secondary Use of Prostate Cancer Real World Data: Automated Classifiers for Clinical and Pathological Stage

Background

Explicit documentation of stage is an endorsed quality metric by the National Quality Forum. Clinical and pathological cancer staging is inconsistently recorded within clinical narratives but can be derived from text in the Electronic Health Record (EHR). To address this need, we developed a Natural Language Processing (NLP) solution for extraction of clinical and pathological TNM stages from the clinical notes in prostate cancer patients.

Methods

Data for patients diagnosed with prostate cancer between 2010 and 2018 were collected from a tertiary care academic healthcare system's EHR records in the United States. This system is linked to the California Cancer Registry, and contains data on diagnosis, histology, cancer stage, treatment and outcomes. A randomly selected sample of patients were manually annotated for stage to establish the ground truth for training and validating the NLP methods. For each patient, a vector representation of clinical text (written in English) was used to train a machine learning model alongside a rule-based model and compared with the ground truth.

Results

A total of 5,461 prostate cancer patients were identified in the clinical data warehouse and over 30% were missing stage information. Thirty-three to thirty-six percent of patients were missing a clinical stage and the models accurately imputed the stage in 21-32% of cases. Twenty-one percent had a missing pathological stage and using NLP 71% of missing T stages and 56% of missing N stages were imputed. For both clinical and pathological T and N stages, the rule-based NLP approach out-performed the ML approach with a minimum F1 score of 0.71 and 0.40, respectively. For clinical M stage the ML approach out-performed the rule-based model with a minimum F1 score of 0.79 and 0.88, respectively.

Conclusions

We developed an NLP pipeline to successfully extract clinical and pathological staging information from clinical narratives. Our results can serve as a proof of concept for using NLP to augment clinical and pathological stage reporting in cancer registries and EHRs to enhance the secondary use of these data.

Artificial Intelligence Applications in Urology: Reporting Standards to Achieve Fluency for Urologists

The growth and adoption of artificial intelligence has led to impressive results in urology. As artificial intelligence grows more ubiquitous, it is important to establish artificial intelligence literacy in the workforce. To this end, we present a narrative review of the literature of artificial intelligence and machine learning in urology and propose a checklist of reporting standards to improve readability and evaluate the current state of the literature. The listed article demonstrated heterogeneous reporting of methodologies and outcomes, limiting generalizability of research. We hope that this review serves as a foundation for future evaluation of medical research in artificial intelligence.

A Novel Multiple Risk Score Model for Prediction of Long-Term Ischemic Risk in Patients With Coronary Artery Disease Undergoing Percutaneous Coronary Intervention: Insights From the I-LOVE-IT 2 Trial

Backgrounds: A plug-and-play standardized algorithm to identify the ischemic risk in patients with coronary artery disease (CAD) undergoing percutaneous coronary intervention (PCI) could play a valuable step to help a wide spectrum of clinic workers. This study intended to investigate the ability to use the accumulation of multiple clinical routine risk scores to predict long-term ischemic events in patients with CAD undergoing PCI.

Methods: This was a secondary analysis of the I-LOVE-IT 2 (Evaluate Safety and Effectiveness of the Tivoli drug-eluting stent (DES) and the Firebird DES for Treatment of Coronary Revascularization) trial, which was a prospective, multicenter, and randomized study. The Global Registry for Acute Coronary Events (GRACE), baseline Synergy Between Percutaneous Coronary Intervention with Taxus and Cardiac Surgery (SYNTAX), residual SYNTAX, and age, creatinine, and ejection fraction (ACEF) score were calculated in all patients. Risk stratification was based on the number of these four scores that met the established thresholds for the ischemic risk. The primary end point was ischemic events at 48 months, defined as the composite of cardiac death, nonfatal myocardial infarction, stroke, or definite/probable stent thrombosis (ST).

Results: The 48-month ischemic events had a significant trend for higher event rates (from 6.61 to 16.93%) with an incremental number of risk scores presenting the higher ischemic risk from 0 to ≥3 (p trend &lt; 0.001). In addition, the categories were associated with increased risk for all components of ischemic events, including cardiac death (from 1.36 to 3.15%), myocardial infarction (MI) (from 3.31 to 9.84%), stroke (3.31 to 6.10%), definite/probable ST (from 0.58 to 1.97%), and all-cause mortality (from 2.14 to 6.30%) (all p trend &lt; 0.05). The net reclassification index after combined with four risk scores was 12.5% (5.3–20.0%), 9.4% (2.0–16.8%), 12.1% (4.5–19.7%), and 10.7% (3.3–18.1%), which offered statistically significant improvement in the performance, compared with SYNTAX, residual SYNTAX, ACEF, and GRACE score, respectively.

Conclusion: The novel multiple risk score model was significantly associated with the risk of long-term ischemic events in these patients with an increment of scores. A meaningful improvement to predict adverse outcomes when multiple risk scores were applied to risk stratification.

Application of Artificial Intelligence to Overcome Clinical Information Overload in Urologic Cancer

OBJECTIVE

To describe the use of artificial intelligence (AI) in medical literature and trial data extraction, and its applications in uro-oncology. This bridging review, which consolidates information from the diverse applications of AI, highlights how AI users can investigate more sophisticated queries than with traditional methods, leading to synthesis of raw data and complex outputs into more actionable and personalized results, particularly in the field of uro-oncology.

METHODS

Literature and clinical trial searches were performed in PubMed, Dimensions, Embase and Google (1999-2020). The searches focused on the use of AI and its various forms to facilitate literature searches, clinical guidelines development, and clinical trial data extraction in uro-oncology. To illustrate how AI can be applied toaddress questions about optimizing therapeutic decision making and individualizing treatment regimens, the Dimensions-linked information platform was searched for "prostate cancer" keywords (76 publications were identified; 48 were included).

RESULTS

AI offers the promise of transforming raw data and complex outputs into actionable insights. Literature and clinical trial searches can be automated, enabling clinicians to develop and analyze publications expeditiously on complex issues such as therapeutic sequencing and to obtain updates on documents that evolve at the pace and scope of the landscape. An AI-based platform inclusive of 12 trial databases and >100 scientific literature sources enabled the creation of an interactive visualization.

CONCLUSION

As the literature and clinical trial landscape continues to grow in complexity and with increasing speed, the ability to pull the right information at the right time from different search engines and resources while excluding social media bias becomes more challenging. This review demonstrates that by applying natural language processing and machine learning algorithms, validated and optimized AI leads to a speedier, more personalized, efficient and focused search compared with traditional methods.

Automated Extraction of Tumor Staging and Diagnosis Information From Surgical Pathology Reports

PURPOSE

Typically stored as unstructured notes, surgical pathology reports contain data elements valuable to cancer research that require labor-intensive manual extraction. Although studies have described natural language processing (NLP) of surgical pathology reports to automate information extraction, efforts have focused on specific cancer subtypes rather than across multiple oncologic domains. To address this gap, we developed and evaluated an NLP method to extract tumor staging and diagnosis information across multiple cancer subtypes.

METHODS

The NLP pipeline was implemented on an open-source framework called Leo. We used a total of 555,681 surgical pathology reports of 329,076 patients to develop the pipeline and evaluated our approach on subsets of reports from patients with breast, prostate, colorectal, and randomly selected cancer subtypes.

RESULTS

Averaged across all four cancer subtypes, the NLP pipeline achieved an accuracy of 1.00 for International Classification of Diseases, Tenth Revision codes, 0.89 for T staging, 0.90 for N staging, and 0.97 for M staging. It achieved an F1 score of 1.00 for International Classification of Diseases, Tenth Revision codes, 0.88 for T staging, 0.90 for N staging, and 0.24 for M staging.

CONCLUSION

The NLP pipeline was developed to extract tumor staging and diagnosis information across multiple cancer subtypes to support the research enterprise in our institution. Although it was not possible to demonstrate generalizability of our NLP pipeline to other institutions, other institutions may find value in adopting a similar NLP approach-and reusing code available at GitHub-to support the oncology research enterprise with elements extracted from surgical pathology reports.

Automatic Classification of Thyroid Findings Using Static and Contextualized Ensemble Natural Language Processing Systems: Development Study

Background

In the case of Korean institutions and enterprises that collect nonstandardized and nonunified formats of electronic medical examination results from multiple medical institutions, a group of experienced nurses who can understand the results and related contexts initially classified the reports manually. The classification guidelines were established by years of workers’ clinical experiences and there were attempts to automate the classification work. However, there have been problems in which rule-based algorithms or human labor–intensive efforts can be time-consuming or limited owing to high potential errors. We investigated natural language processing (NLP) architectures and proposed ensemble models to create automated classifiers.

Objective

This study aimed to develop practical deep learning models with electronic medical records from 284 health care institutions and open-source corpus data sets for automatically classifying 3 thyroid conditions: healthy, caution required, and critical. The primary goal is to increase the overall accuracy of the classification, yet there are practical and industrial needs to correctly predict healthy (negative) thyroid condition data, which are mostly medical examination results, and minimize false-negative rates under the prediction of healthy thyroid conditions.

Methods

The data sets included thyroid and comprehensive medical examination reports. The textual data are not only documented in fully complete sentences but also written in lists of words or phrases. Therefore, we propose static and contextualized ensemble NLP network (SCENT) systems to successfully reflect static and contextual information and handle incomplete sentences. We prepared each convolution neural network (CNN)-, long short-term memory (LSTM)-, and efficiently learning an encoder that classifies token replacements accurately (ELECTRA)-based ensemble model by training or fine-tuning them multiple times. Through comprehensive experiments, we propose 2 versions of ensemble models, SCENT-v1 and SCENT-v2, with the single-architecture–based CNN, LSTM, and ELECTRA ensemble models for the best classification performance and practical use, respectively. SCENT-v1 is an ensemble of CNN and ELECTRA ensemble models, and SCENT-v2 is a hierarchical ensemble of CNN, LSTM, and ELECTRA ensemble models. SCENT-v2 first classifies the 3 labels using an ELECTRA ensemble model and then reclassifies them using an ensemble model of CNN and LSTM if the ELECTRA ensemble model predicted them as “healthy” labels.

Results

SCENT-v1 outperformed all the suggested models, with the highest F1 score (92.56%). SCENT-v2 had the second-highest recall value (94.44%) and the fewest misclassifications for caution-required thyroid condition while maintaining 0 classification error for the critical thyroid condition under the prediction of the healthy thyroid condition.

Conclusions

The proposed SCENT demonstrates good classification performance despite the unique characteristics of the Korean language and problems of data lack and imbalance, especially for the extremely low amount of critical condition data. The result of SCENT-v1 indicates that different perspectives of static and contextual input token representations can enhance classification performance. SCENT-v2 has a strong impact on the prediction of healthy thyroid conditions.

Natural Language Processing to Identify Cancer Treatments With Electronic Medical Records

PURPOSE

Knowing the treatments administered to patients with cancer is important for treatment planning and correlating treatment patterns with outcomes for personalized medicine study. However, existing methods to identify treatments are often lacking. We develop a natural language processing approach with structured electronic medical records and unstructured clinical notes to identify the initial treatment administered to patients with cancer.

METHODS

We used a total number of 4,412 patients with 483,782 clinical notes from the Stanford Cancer Institute Research Database containing patients with nonmetastatic prostate, oropharynx, and esophagus cancer. We trained treatment identification models for each cancer type separately and compared performance of using only structured, only unstructured (bag-of-words, doc2vec, fasttext), and combinations of both (structured + bow, structured + doc2vec, structured + fasttext). We optimized the identification model among five machine learning methods (logistic regression, multilayer perceptrons, random forest, support vector machines, and stochastic gradient boosting). The treatment information recorded in the cancer registry is the gold standard and compares our methods to an identification baseline with billing codes.

RESULTS

For prostate cancer, we achieved an f1-score of 0.99 (95% CI, 0.97 to 1.00) for radiation and 1.00 (95% CI, 0.99 to 1.00) for surgery using structured + doc2vec. For oropharynx cancer, we achieved an f1-score of 0.78 (95% CI, 0.58 to 0.93) for chemoradiation and 0.83 (95% CI, 0.69 to 0.95) for surgery using doc2vec. For esophagus cancer, we achieved an f1-score of 1.0 (95% CI, 1.0 to 1.0) for both chemoradiation and surgery using all combinations of structured and unstructured data. We found that employing the free-text clinical notes outperforms using the billing codes or only structured data for all three cancer types.

CONCLUSION

Our results show that treatment identification using free-text clinical notes greatly improves upon the performance using billing codes and simple structured data. The approach can be used for treatment cohort identification and adapted for longitudinal cancer treatment identification.

Accurate pattern-based extraction of complex Gleason score expressions from pathology reports

PURPOSE

The Gleason score is an important grading factor of prostate cancer. Gleason scores can be extracted from pathology report texts using regular expressions, but previously developed programmes have targeted only relatively simple Gleason score expressions. We developed a programme capable of extracting also complex expressions. The programme is relatively easy to adapt to other languages and datasets.

METHODS

We developed and evaluated our regular expression-based programme using manually processed pathology reports of prostate cancer cases diagnosed in Finland in 2016-2017. Both simple and complex Gleason score expressions were targeted. We measured the performance of our programme using recall, precision, and the F1. The proportion of complex Gleason score expressions was estimated as the complement of the recall when only addition expressions (e.g. "Gleason 3 + 4") were targeted.

RESULTS

The detection of values (scores and score components) is based on mandatory keywords before or after the value. The programme favours precision over recall by primarily allowing for lists of optional expressions between keyword-value pairs and only secondarily allowing for arbitrary expressions. The programme is straightforward to adapt to new datasets by modifying the lists of mandatory and optional expressions. The full and addition-only programmes had 92% (95% CI: [90%, 95%]) and 65% ([61%, 70%]) recall and high precision (98% [97%, 99%] and 100% [99%, 100%]), respectively. The estimated proportion of complex Gleason score expressions was 100-65 = 35%.

CONCLUSIONS

Even complex Gleason score expressions can be extracted with high recall and precision using regular expressions. We recommend implementing automated Gleason score extraction where possible by adapting our validated programme.

Collecting specialty-related medical terms: Development and evaluation of a resource for Spanish

BACKGROUND

Controlled vocabularies are fundamental resources for information extraction from clinical texts using natural language processing (NLP). Standard language resources available in the healthcare domain such as the UMLS metathesaurus or SNOMED CT are widely used for this purpose, but with limitations such as lexical ambiguity of clinical terms. However, most of them are unambiguous within text limited to a given clinical specialty. This is one rationale besides others to classify clinical text by the clinical specialty to which they belong.

RESULTS

This paper addresses this limitation by proposing and applying a method that automatically extracts Spanish medical terms classified and weighted per sub-domain, using Spanish MEDLINE titles and abstracts as input. The hypothesis is biomedical NLP tasks benefit from collections of domain terms that are specific to clinical subdomains. We use PubMed queries that generate sub-domain specific corpora from Spanish titles and abstracts, from which token n-grams are collected and metrics of relevance, discriminatory power, and broadness per sub-domain are computed. The generated term set, called Spanish core vocabulary about clinical specialties (SCOVACLIS), was made available to the scientific community and used in a text classification problem obtaining improvements of 6 percentage points in the F-measure compared to the baseline using Multilayer Perceptron, thus demonstrating the hypothesis that a specialized term set improves NLP tasks.

CONCLUSION

The creation and validation of SCOVACLIS support the hypothesis that specific term sets reduce the level of ambiguity when compared to a specialty-independent and broad-scope vocabulary.

Validation of deep learning natural language processing algorithm for keyword extraction from pathology reports in electronic health records

Pathology reports contain the essential data for both clinical and research purposes. However, the extraction of meaningful, qualitative data from the original document is difficult due to the narrative and complex nature of such reports. Keyword extraction for pathology reports is necessary to summarize the informative text and reduce intensive time consumption. In this study, we employed a deep learning model for the natural language process to extract keywords from pathology reports and presented the supervised keyword extraction algorithm. We considered three types of pathological keywords, namely specimen, procedure, and pathology types. We compared the performance of the present algorithm with the conventional keyword extraction methods on the 3115 pathology reports that were manually labeled by professional pathologists. Additionally, we applied the present algorithm to 36,014 unlabeled pathology reports and analysed the extracted keywords with biomedical vocabulary sets. The results demonstrated the suitability of our model for practical application in extracting important data from pathology reports.

Using Systematized Nomenclature of Medicine clinical term codes to assign histological findings for prostate biopsies in the Gauteng province, South Africa: Lessons learnt

Background

Prostate cancer (PCa) is a leading male neoplasm in South Africa.

Objective

The aim of our study was to describe PCa using Systemized Nomenclature of Medicine (SNOMED) clinical terms codes, which have the potential to generate more timely data.

Methods

The retrospective study design was used to analyse prostate biopsy data from our laboratories using SNOMED morphology (M) and topography (T) codes where the term ’prostate’ was captured in the narrative report. Using M code descriptions, the diagnosis, sub-diagnosis, sub-result and International Classification of Diseases for Oncology (ICD-O-3) codes were assigned using a lookup table. Topography code descriptions identified biopsies of prostatic origin. Lookup tables were prepared using Microsoft Excel and combined with the data extracts using Access. Contingency tables reported M and T codes, diagnosis and sub-diagnosis frequencies.

Results

An M and T code was reported for 88% (n = 22 009) of biopsies. Of these, 20 551 (93.37%) were of prostatic origin. A benign diagnosis (ICD-O-3:8000/0) was reported for 10 441 biopsies (50.81%) and 45.26% had a malignant diagnosis (n = 9302). An adenocarcinoma (8140/3) sub-diagnosis was reported for 88.16% of malignant biopsies (n = 8201). An atypia diagnosis was reported for 760 biopsies (3.7%). Inflammation (39.03%) and hyperplasia (20.82%) were the predominant benign sub-diagnoses.

Conclusion

Our study demonstrated the feasibility of generating PCa data using SNOMED codes from national laboratory data. This highlights the need for extending the results of our study to a national level to deliver timeous monitoring of PCa trends.

Obtaining Knowledge in Pathology Reports Through a Natural Language Processing Approach With Classification, Named-Entity Recognition, and Relation-Extraction Heuristics

PURPOSE

Robust institutional tumor banks depend on continuous sample curation or else subsequent biopsy or resection specimens are overlooked after initial enrollment. Curation automation is hindered by semistructured free-text clinical pathology notes, which complicate data abstraction. Our motivation is to develop a natural language processing method that dynamically identifies existing pathology specimen elements necessary for locating specimens for future use in a manner that can be re-implemented by other institutions.

PATIENTS AND METHODS

Pathology reports from patients with gastroesophageal cancer enrolled in The University of Chicago GI oncology tumor bank were used to train and validate a novel composite natural language processing-based pipeline with a supervised machine learning classification step to separate notes into internal (primary review) and external (consultation) reports; a named-entity recognition step to obtain label (accession number), location, date, and sublabels (block identifiers); and a results proofreading step.

RESULTS

We analyzed 188 pathology reports, including 82 internal reports and 106 external consult reports, and successfully extracted named entities grouped as sample information (label, date, location). Our approach identified up to 24 additional unique samples in external consult notes that could have been overlooked. Our classification model obtained 100% accuracy on the basis of 10-fold cross-validation. Precision, recall, and F1 for class-specific named-entity recognition models show strong performance.

CONCLUSION

Through a combination of natural language processing and machine learning, we devised a re-implementable and automated approach that can accurately extract specimen attributes from semistructured pathology notes to dynamically populate a tumor registry.

A Novel Approach for Fully Automatic Intra-Tumor Segmentation With 3D U-Net Architecture for Gliomas

Purpose: Gliomas are the most common primary brain malignancies, with varying degrees of aggressiveness and prognosis. Understanding of tumor biology and intra-tumor heterogeneity is necessary for planning personalized therapy and predicting response to therapy. Accurate tumoral and intra-tumoral segmentation on MRI is the first step toward understanding the tumor biology through computational methods. The purpose of this study was to design a segmentation algorithm and evaluate its performance on pre-treatment brain MRIs obtained from patients with gliomas.

Materials and Methods: In this study, we have designed a novel 3D U-Net architecture that segments various radiologically identifiable sub-regions like edema, enhancing tumor, and necrosis. Weighted patch extraction scheme from the tumor border regions is proposed to address the problem of class imbalance between tumor and non-tumorous patches. The architecture consists of a contracting path to capture context and the symmetric expanding path that enables precise localization. The Deep Convolutional Neural Network (DCNN) based architecture is trained on 285 patients, validated on 66 patients and tested on 191 patients with Glioma from Brain Tumor Segmentation (BraTS) 2018 challenge dataset. Three dimensional patches are extracted from multi-channel BraTS training dataset to train 3D U-Net architecture. The efficacy of the proposed approach is also tested on an independent dataset of 40 patients with High Grade Glioma from our tertiary cancer center. Segmentation results are assessed in terms of Dice Score, Sensitivity, Specificity, and Hausdorff 95 distance (ITCN intra-tumoral classification network).

Result: Our proposed architecture achieved Dice scores of 0.88, 0.83, and 0.75 for the whole tumor, tumor core and enhancing tumor, respectively, on BraTS validation dataset and 0.85, 0.77, 0.67 on test dataset. The results were similar on the independent patients' dataset from our hospital, achieving Dice scores of 0.92, 0.90, and 0.81 for the whole tumor, tumor core and enhancing tumor, respectively.

Conclusion: The results of this study show the potential of patch-based 3D U-Net for the accurate intra-tumor segmentation. From experiments, it is observed that the weighted patch-based segmentation approach gives comparable performance with the pixel-based approach when there is a thin boundary between tumor subparts.