INTRODUCTION: Machine learning algorithms are expected to work side-by-side with humans in decision-making pipelines. Thus, the ability of classifiers to make reliable decisions is of paramount importance. Deep neural networks (DNNs) represent the state-of-the-art models to address real-world classification. Although the strength of activation in DNNs is often correlated with the network's confidence, in-depth analyses are needed to establish whether they are well calibrated. METHOD: In this paper, we demonstrate the use of DNN-based classification tools to benefit cancer registries by automating information extraction of disease at diagnosis and at surgery from electronic text pathology reports from the US National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) population-based cancer registries. In particular, we introduce multiple methods for selective classification to achieve a target level of accuracy on multiple classification tasks while minimizing the rejection amount-that is, the number of electronic pathology reports for which the model's predictions are unreliable. We evaluate the proposed methods by comparing our approach with the current in-house deep learning-based abstaining classifier. RESULTS: Overall, all the proposed selective classification methods effectively allow for achieving the targeted level of accuracy or higher in a trade-off analysis aimed to minimize the rejection rate. On in-distribution validation and holdout test data, with all the proposed methods, we achieve on all tasks the required target level of accuracy with a lower rejection rate than the deep abstaining classifier (DAC). Interpreting the results for the out-of-distribution test data is more complex; nevertheless, in this case as well, the rejection rate from the best among the proposed methods achieving 97% accuracy or higher is lower than the rejection rate based on the DAC. CONCLUSIONS: We show that although both approaches can flag those samples that should be manually reviewed and labeled by human annotators, the newly proposed methods retain a larger fraction and do so without retraining-thus offering a reduced computational cost compared with the in-house deep learning-based abstaining classifier.

In the last decade, the widespread adoption of electronic health record documentation has created huge opportunities for information mining. Natural language processing (NLP) techniques using machine and deep learning are becoming increasingly widespread for information extraction tasks from unstructured clinical notes. Disparities in performance when deploying machine learning models in the real world have recently received considerable attention. In the clinical NLP domain, the robustness of convolutional neural networks (CNNs) for classifying cancer pathology reports under natural distribution shifts remains understudied. In this research, we aim to quantify and improve the performance of the CNN for text classification on out-of-distribution (OOD) datasets resulting from the natural evolution of clinical text in pathology reports. We identified class imbalance due to different prevalence of cancer types as one of the sources of performance drop and analyzed the impact of previous methods for addressing class imbalance when deploying models in real-world domains. Our results show that our novel class-specialized ensemble technique outperforms other methods for the classification of rare cancer types in terms of macro F1 scores. We also found that traditional ensemble methods perform better in top classes, leading to higher micro F1 scores. Based on our findings, we formulate a series of recommendations for other ML practitioners on how to build robust models with extremely imbalanced datasets in biomedical NLP applications.

Recent applications ofdeep learning have shown promising results for classifying unstructured text in the healthcare domain. However, the reliability of models in production settings has been hindered by imbalanced data sets in which a small subset of the classes dominate. In the absence of adequate training data, rare classes necessitate additional model constraints for robust performance. Here, we present a strategy for incorporating short sequences of text (i.e. keywords) into training to boost model accuracy on rare classes. In our approach, we assemble a set of keywords, including short phrases, associated with each class. The keywords are then used as additional data during each batch of model training, resulting in a training loss that has contributions from both raw data and keywords. We evaluate our approach on classification of cancer pathology reports, which shows a substantial increase in model performance for rare classes. Furthermore, we analyze the impact of keywords on model output probabilities for bigrams, providing a straightforward method to identify model difficulties for limited training data.

BACKGROUND: Automated text classification has many important applications in the clinical setting; however, obtaining labelled data for training machine learning and deep learning models is often difficult and expensive. Active learning techniques may mitigate this challenge by reducing the amount of labelled data required to effectively train a model. In this study, we analyze the effectiveness of 11 active learning algorithms on classifying subsite and histology from cancer pathology reports using a Convolutional Neural Network as the text classification model. RESULTS: We compare the performance of each active learning strategy using two differently sized datasets and two different classification tasks. Our results show that on all tasks and dataset sizes, all active learning strategies except diversity-sampling strategies outperformed random sampling, i.e., no active learning. On our large dataset (15K initial labelled samples, adding 15K additional labelled samples each iteration of active learning), there was no clear winner between the different active learning strategies. On our small dataset (1K initial labelled samples, adding 1K additional labelled samples each iteration of active learning), marginal and ratio uncertainty sampling performed better than all other active learning techniques. We found that compared to random sampling, active learning strongly helps performance on rare classes by focusing on underrepresented classes. CONCLUSIONS: Active learning can save annotation cost by helping human annotators efficiently and intelligently select which samples to label. Our results show that a dataset constructed using effective active learning techniques requires less than half the amount of labelled data to achieve the same performance as a dataset constructed using random sampling.

Bidirectional Encoder Representations from Transformers (BERT) and BERT-based approaches are the current state-of-the-art in many natural language processing (NLP) tasks; however, their application to document classification on long clinical texts is limited. In this work, we introduce four methods to scale BERT, which by default can only handle input sequences up to approximately 400 words long, to perform document classification on clinical texts several thousand words long. We compare these methods against two much simpler architectures - a word-level convolutional neural network and a hierarchical self-attention network - and show that BERT often cannot beat these simpler baselines when classifying MIMIC-III discharge summaries and SEER cancer pathology reports. In our analysis, we show that two key components of BERT - pretraining and WordPiece tokenization - may actually be inhibiting BERT's performance on clinical text classification tasks where the input document is several thousand words long and where correctly identifying labels may depend more on identifying a few key words or phrases rather than understanding the contextual meaning of sequences of text.

OBJECTIVE: We implement 2 different multitask learning (MTL) techniques, hard parameter sharing and cross-stitch, to train a word-level convolutional neural network (CNN) specifically designed for automatic extraction of cancer data from unstructured text in pathology reports. We show the importance of learning related information extraction (IE) tasks leveraging shared representations across the tasks to achieve state-of-the-art performance in classification accuracy and computational efficiency. MATERIALS AND METHODS: Multitask CNN (MTCNN) attempts to tackle document information extraction by learning to extract multiple key cancer characteristics simultaneously. We trained our MTCNN to perform 5 information extraction tasks: (1) primary cancer site (65 classes), (2) laterality (4 classes), (3) behavior (3 classes), (4) histological type (63 classes), and (5) histological grade (5 classes). We evaluated the performance on a corpus of 95 231 pathology documents (71 223 unique tumors) obtained from the Louisiana Tumor Registry. We compared the performance of the MTCNN models against single-task CNN models and 2 traditional machine learning approaches, namely support vector machine (SVM) and random forest classifier (RFC). RESULTS: MTCNNs offered superior performance across all 5 tasks in terms of classification accuracy as compared with the other machine learning models. Based on retrospective evaluation, the hard parameter sharing and cross-stitch MTCNN models correctly classified 59.04% and 57.93% of the pathology reports respectively across all 5 tasks. The baseline models achieved 53.68% (CNN), 46.37% (RFC), and 36.75% (SVM). Based on prospective evaluation, the percentages of correctly classified cases across the 5 tasks were 60.11% (hard parameter sharing), 58.13% (cross-stitch), 51.30% (single-task CNN), 42.07% (RFC), and 35.16% (SVM). Moreover, hard parameter sharing MTCNNs outperformed the other models in computational efficiency by using about the same number of trainable parameters as a single-task CNN. CONCLUSIONS: The hard parameter sharing MTCNN offers superior classification accuracy for automated coding support of pathology documents across a wide range of cancers and multiple information extraction tasks while maintaining similar training and inference time as those of a single task-specific model.

Cancer registries collect unstructured and structured cancer data for surveillance purposes which provide important insights regarding cancer characteristics, treatments, and outcomes. Cancer registry data typically (1) categorize each reportable cancer case or tumor at the time of diagnosis, (2) contain demographic information about the patient such as age, gender, and location at time of diagnosis, (3) include planned and completed primary treatment information, and (4) may contain survival outcomes. As structured data is being extracted from various unstructured sources, such as pathology reports, radiology reports, medical records, and stored for reporting and other needs, the associated information representing a reportable cancer is constantly expanding and evolving. While some popular analytic approaches including SEER*Stat and SAS exist, we provide a knowledge graph approach to organizing cancer registry data. Our approach offers unique advantages for timely data analysis and presentation and visualization of valuable information. This knowledge graph approach semantically enriches the data, and easily enables linking with third-party data which can help explain variation in cancer incidence patterns, disparities, and outcomes. We developed a prototype knowledge graph based on the Louisiana Tumor Registry dataset. We present the advantages of the knowledge graph approach by examining: i) scenario-specific queries, ii) links with openly available external datasets, iii) schema evolution for iterative analysis, and iv) data visualization. Our results demonstrate that this graph based solution can perform complex queries, improve query run-time performance by up to 76%, and more easily conduct iterative analyses to enhance researchers' understanding of cancer registry data.

Collaboration among cancer registries is essential to develop accurate, robust, and generalizable deep learning models for automated information extraction from cancer pathology reports. Sharing data presents a serious privacy issue, especially in biomedical research and healthcare delivery domains. Distributing pretrained deep learning (DL) models has been proposed to avoid critical data sharing. However, there is growing recognition that collaboration among clinical institutes through DL model distribution exposes new security and privacy vulnerabilities. These vulnerabilities increase in natural language processing (NLP) applications, in which the dataset vocabulary with word vector representations needs to be associated with the other model parameters. In this paper, we propose a novel privacy-preserving DL model distribution across cancer registries for information extraction from cancer pathology reports with privacy and confidentiality considerations. The proposed approach exploits the adversarial training framework to distinguish private features from shared features among different datasets. It only shares registry-invariant model parameters, without sharing raw data nor registry-specific model parameters among cancer registries. Thus, it protects both the data and the trained model simultaneously. We compare our proposed approach to single-registry models, and a model trained on centrally hosted data from different cancer registries. The results show that the proposed approach significantly outperforms the single-registry models and achieves statistically indistinguishable micro and macro F1-score as compared to the centralized model.

We introduce a deep learning architecture, hierarchical self-attention networks (HiSANs), designed for classifying pathology reports and show how its unique architecture leads to a new state-of-the-art in accuracy, faster training, and clear interpretability. We evaluate performance on a corpus of 374,899 pathology reports obtained from the National Cancer Institute's (NCI) Surveillance, Epidemiology, and End Results (SEER) program. Each pathology report is associated with five clinical classification tasks - site, laterality, behavior, histology, and grade. We compare the performance of the HiSAN against other machine learning and deep learning approaches commonly used on medical text data - Naive Bayes, logistic regression, convolutional neural networks, and hierarchical attention networks (the previous state-of-the-art). We show that HiSANs are superior to other machine learning and deep learning text classifiers in both accuracy and macro F-score across all five classification tasks. Compared to the previous state-of-the-art, hierarchical attention networks, HiSANs not only are an order of magnitude faster to train, but also achieve about 1% better relative accuracy and 5% better relative macro F-score.

BACKGROUND: Manual extraction of information from electronic pathology (epath) reports to populate the Surveillance, Epidemiology, and End Result (SEER) database is labor intensive. Systematizing the data extraction automatically using machine-learning (ML) and natural language processing (NLP) is desirable to reduce the human labor required to populate the SEER database and to improve the timeliness of the data. This enables scaling up registry efficiency and collection of new data elements. To ensure the integrity, quality, and continuity of the SEER data, the misclassification error of ML and NPL algorithms needs to be negligible. Current algorithms fail to achieve the precision of human experts who can bring additional information in their assessments. Differences in registry format and the desire to develop a common information extraction platform further complicate the ML/NLP tasks. The purpose of our study is to develop triage rules to partially automate registry workflow to improve the precision of the auto-extracted information. RESULTS: This paper presents a mathematical framework to improve the precision of a classifier beyond that of the Bayes classifier by selectively classifying item that are most likely to be correct. This results in a triage rule that only classifies a subset of the item. We characterize the optimal triage rule and demonstrate its usefulness in the problem of classifying cancer site from electronic pathology reports to achieve a desired precision. CONCLUSIONS: From the mathematical formalism, we propose a heuristic estimate for triage rule based on post-processing the soft-max output from standard machine learning algorithms. We show, in test cases, that the triage rule significantly improve the classification accuracy.

BACKGROUND: Deep Learning (DL) has advanced the state-of-the-art capabilities in bioinformatics applications which has resulted in trends of increasingly sophisticated and computationally demanding models trained by larger and larger data sets. This vastly increased computational demand challenges the feasibility of conducting cutting-edge research. One solution is to distribute the vast computational workload across multiple computing cluster nodes with data parallelism algorithms. In this study, we used a High-Performance Computing environment and implemented the Downpour Stochastic Gradient Descent algorithm for data parallelism to train a Convolutional Neural Network (CNN) for the natural language processing task of information extraction from a massive dataset of cancer pathology reports. We evaluated the scalability improvements using data parallelism training and the Titan supercomputer at Oak Ridge Leadership Computing Facility. To evaluate scalability, we used different numbers of worker nodes and performed a set of experiments comparing the effects of different training batch sizes and optimizer functions. RESULTS: We found that Adadelta would consistently converge at a lower validation loss, though requiring over twice as many training epochs as the fastest converging optimizer, RMSProp. The Adam optimizer consistently achieved a close 2nd place minimum validation loss significantly faster; using a batch size of 16 and 32 allowed the network to converge in only 4.5 training epochs. CONCLUSIONS: We demonstrated that the networked training process is scalable across multiple compute nodes communicating with message passing interface while achieving higher classification accuracy compared to a traditional machine learning algorithm.