Author:
de Oliveira Ricardo,Menezes Bruno,Ortiz Júnia,Nascimento Erick
Abstract
This chapter aims to present a classification model for categorizing textual clinical records of breast magnetic resonance imaging, based on lexical, syntactic and semantic analysis of clinical reports according to the Breast Imaging-Reporting and Data System (BI-RADS) classification, using Deep Learning and Natural Language Processing (NLP). The model was developed from transfer learning based on the pre-trained BERTimbau model, BERT model (Bidirectional Encoder Representations from Transformers) trained in Brazilian Portuguese. The dataset is composed of medical reports in Brazilian Portuguese classified into six categories: Inconclusive; Normal or Negative; Certainly Benign Findings; Probably Benign Findings; Suspicious Findings; High Risk of Cancer; Previously Known Malignant Injury. The following models were implemented and compared: Random Forest, SVM, Naïve Bayes, BERTimbau with and without finetuning. The BERTimbau model presented better results, with better performance after finetuning.
Reference14 articles.
1. Castro S, M, Tseytlin E, Medvedeva O, Mitchhell K, Visweswaran S, Bekhuis T, et al. Automated annotation and classification of BI-RADS assessment from radiology reports. Journal of Biomedical Informatics. 2017
2. Souza F, Nogueira R, Lotufo R. BERTimbau: Pretrained BERT models for Brazilian Portuguese. In: 9th Brazilian Conference on Intelligent Systems. Rio Grande do Sul, Brazil, October 20-23 (to appear). [S.l.: s.n.]: BRACIS; 2020
3. Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser L, Polosukhin I. Available from: https://proceedings.neurips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf
4. Devlin J, Chang MW, Lee K, Toutanova K. BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding. 2019. Available from: https://arxiv.org/abs/1810.04805
5. Dai Andrew M, Le Quoc V. Semi-Supervised Sequence Learning. Available from: https://arxiv.org/abs/1511.01432. 2015