Cancer Document Classification Leveraging Graph-Based Residual Network for Scenarios with Limited Data

PDF view of the paper entitled Medical-Gat: Classification of the Cancer Document. Take advantage of the remaining network based on the script graph with limited data, by Elias Hussein and 3 other authors

PDF HTML (experimental) view

a summary:The exact classification of cancer -related medical summary is crucial for healthcare and research management. However, obtaining large data groups called in the medical field is difficult due to privacy concerns and clinical data complexity. This is the scarcity of explanatory data that hinders the development of effective automatic learning models to classify the cancer document. To face this challenge, we offer a set of coordinated data that includes 1,874 vital medical summaries, classified as thyroid cancer, colon cancer, lung cancer and public topics. Our research focuses on taking advantage of this data set to improve classification performance, especially in data scope scenarios. We offer the R-GAT attention network with the multiple graphic attention layers that capture significant information and structural relationships within the documents related to cancer. Our R-GAT model is compared to various technologies, including transformer-based models such as Transformers (BERT), Roberta, and field models such as BIOBERT and BIO+ClinicalBert. We also evaluated the deep learning models (CNNS, LSTMS) and traditional automatic learning models (logistical slope, SVM). In addition, we explore the band’s approaches that combine deep learning models to enhance classification. Various features are evaluated, including the frequency of the frequency document (TF-IDF) with Unigrams and Bigrams, Word2VEC, and the features of BERT and Roberta. R-GAT model outperforms other technologies, achieve accuracy, summons, F1 grades of 0.99, 0.97, and 0.98 thyroid cancer; 0.96, 0.94, and 0.95 colon cancer; 0.96, 0.99, and 0.97 lung cancer; 0.95, 0.96, and 0.95 for public topics.