A Systematic Review of Application Progress on Machine Learning-Based Natural Language Processing in Breast Cancer over the Past 5 Years

Chengtai Li, Ying Weng, Yiming Zhang, Boding WangDiagnostics 2023 · DOI

TL;DR

This systematic review analyzes the application of machine learning-based Natural Language Processing (NLP) in breast cancer diagnosis over the last five years. It categorizes and evaluates various NLP models, identifies key challenges like data scarcity, and proposes future research directions to improve AI's role in breast cancer care.

What Problem Does This Paper Address?

Breast cancer remains a significant global health concern, and while AI, particularly machine learning (ML), shows immense promise in improving diagnosis and efficiency, its application in processing the vast amount of medical text data is still evolving. Doctors and researchers often face a steep learning curve when navigating the complex landscape of ML-based NLP models. This paper aims to bridge this gap by providing a comprehensive overview of the current state of NLP applications in breast cancer, detailing the models used, their strengths and weaknesses, and the challenges hindering progress, thereby guiding future research and clinical adoption.

Methods

The authors conducted a systematic review following the PRISMA guidelines, searching major databases (PubMed, Google Scholar, Web of Science, Scopus, ACL Anthology) for literature published between 2018 and 2022. They screened over 200 publications, ultimately including 25 papers that focused on ML-based NLP applications in breast cancer. The review categorizes ML models into two main groups: conventional ML (e.g., CRF, SVM, K-means) and deep learning (DL) models (e.g., LSTM, BERT, CNN). For each model, they discuss its theoretical foundation, advantages, disadvantages, and specific applications in breast cancer research.

Datasets

The paper highlights the critical role of datasets in NLP applications for breast cancer. It notes that while online forums provide large datasets (e.g., for sentiment analysis), medical text datasets are often smaller and more specialized due to the technical nature of the data and the need for expert annotation. Many datasets used in the reviewed studies are private, posing challenges for reproducibility and generalizability. Publicly available datasets mentioned include:

  • Breast Cancer Coimbra Dataset
  • Blog Articles on Life Palette
  • Tweets from Twitter
  • Text from MedHelp
  • Oncoshare Breast Cancer Database (partially available)
  • I2B2 NLP Research Database
  • MIMIC-III Critical Care Database
  • eDiseases Dataset
  • China Medical University Hospital (CMUH) database (available on request)

Key Insights and Ideas

  • Shift Towards Deep Learning: The review clearly indicates a growing trend towards deep learning models (especially BERT and LSTM) over conventional ML models in recent years, reflecting the broader AI landscape's evolution towards more complex architectures for high-precision tasks.
  • The Data Bottleneck: A significant challenge identified is the scarcity of large, publicly available, and well-annotated datasets for breast cancer NLP tasks. This limitation impacts model development, validation, and comparison.
  • Bridging the Gap: The paper emphasizes the need to close the research gap between AI experts and medical professionals, making complex NLP models more accessible and understandable to clinicians.
  • Model Versatility: Different NLP models are suited for different tasks. For instance, CRF is useful for serial data and limited resources, while BERT excels at capturing long-range contextual relationships.

Key Results

The review identified 25 relevant papers, categorizing the ML models used into conventional (CRF, SVM, K-means) and deep learning (LSTM, BERT, CNN). The trend analysis shows a clear increase in the application of DL models, particularly BERT, from 2018 to 2022, with conventional models being gradually replaced. The paper details the strengths and weaknesses of each model type in the context of breast cancer NLP, such as CRF's suitability for serial data and limited resources, SVM's effectiveness on small datasets, LSTM's ability to handle sequential data, BERT's superior contextual understanding, and CNN's computational efficiency.

Limitations and Future Work

The authors acknowledge several limitations:

  • Dataset Scarcity: The primary challenge is the lack of adequate public datasets, leading to reliance on private datasets with potential biases and quality issues. This limits the generalizability of models and hinders direct comparison of experimental results.
  • Model Interpretability: Deep learning models, while powerful, often suffer from a "black box" nature, making their results difficult to interpret for clinicians.

Future work directions suggested to address these challenges include:

  • Semi-Supervised Learning: Leveraging unlabeled data to improve model performance when labeled data is scarce.
  • Active Learning: Strategically selecting data points for annotation to maximize model improvement with minimal labeling effort.
  • Transfer Learning: Adapting models trained on large datasets from related domains to breast cancer-specific tasks, reducing the need for extensive target-domain data.

Summary updated

Abstract

Artificial intelligence (AI) has been steadily developing in the medical field in the past few years, and AI-based applications have advanced cancer diagnosis. Breast cancer has a massive amount of data in oncology. There has been a high level of research enthusiasm to apply AI techniques to assist in breast cancer diagnosis and improve doctors' efficiency. However, the wise utilization of tedious breast cancer-related medical care is still challenging. Over the past few years, AI-based NLP applications have been increasingly proposed in breast cancer. In this systematic review, we conduct the review using preferred reporting items for systematic reviews and meta-analyses (PRISMA) and investigate the recent five years of literature in natural language processing (NLP)-based AI applications. This systematic review aims to uncover the recent trends in this area, close the research gap, and help doctors better understand the NLP application pipeline. We first conduct an initial literature search of 202 publications from Scopus, Web of Science, PubMed, Google Scholar, and the Association for Computational Linguistics (ACL) Anthology. Then, we screen the literature based on inclusion and exclusion criteria. Next, we categorize and analyze the advantages and disadvantages of the different machine learning models. We also discuss the current challenges, such as the lack of a public dataset. Furthermore, we suggest some promising future directions, including semi-supervised learning, active learning, and transfer learning.
BibTeX
@inproceedings{li2023systematic,
  author = {Chengtai Li and Ying Weng and Yiming Zhang and Boding Wang},
  title = {{A Systematic Review of Application Progress on Machine Learning-Based Natural Language Processing in Breast Cancer over the Past 5 Years}},
  booktitle = {Diagnostics},
  year = {2023},
  doi = {10.3390/diagnostics13030537},
}
Figure