Learning to ground medical text in a 3D human atlas

Proceedings of the 24th Conference on Computational Natural Language Learning, pages 302–312 Online, November 19-20, 2020. c⃝2020 Association for Computational Linguistics https://doi.org/10.18653/v1/P17 302 Learning to Ground Medical Text in a 3D Human Atlas Dusan Grujicic∗, Gorjan Radevski∗, Tinne Tuytelaars, Matthew B. Blaschko Department of Electrical Engineering (ESAT-PSI) KU Leuven firstname.lastname@esat.kuleuven.be Abstract In this paper, we develop a method for ground- ing medical text into a physically meaning- ful and interpretable space corresponding to a human atlas. We build on text embedding architectures such as BERT and introduce a loss function that allows us to reason about the semantic and spatial relatedness of medi- cal texts by learning a projection of the em- bedding into a 3D space representing the hu- man body. We quantitatively and qualitatively demonstrate that our proposed method learns a context sensitive and spatially aware mapping, in both the inter-organ and intra-organ sense, using a large scale medical text dataset from the “Large-scale online biomedical semantic indexing” track of the 2020 BioASQ challenge. We extend our approach to a self-supervised setting, and find it to be competitive with a classification based method, and a fully super- vised variant of approach. 1 Introduction The quantity of available medical literature in- creases daily (Wang et al.; Tsatsaronis et al., 2015), however, it is often provided in a non-systematized, free form. The development of BERT (Devlin et al., 2018), and the increased popularity of trans- fer learning in natural language processing (NLP), prompted notable works that aim to leverage pub- licly available medical and scientific articles to de- velop domain specific pre-trained language models (Lee et al., 2019; Alsentzer et al., 2019; Beltagy et al., 2019; Jin et al., 2019). High quality sentence representations that capture the semantics and struc- ture of the text can be obtained by training models to solve the Natural Language Inference (NLI) task on open-domain datasets (Bowman et al., 2015; Williams et al., 2017) and predict whether two pieces of text entail, contradict or are neutral to ∗Equal Contribution Figure 1: Given the text with implicit refer- ence to lungs: “Divided into two lobes, an up- per and a lower lobe, by the oblique fissure, which extends from the costal to the mediastinal surface” (Drake et al., 2009), our model learns the grounding indicated by the star. each other (Conneau et al., 2017). The aforemen- tioned BERT-based models can serve as the en- coder backbone for such approaches (Reimers and Gurevych, 2019), and the setup can be trivially ex- tended to enable searching through and retrieving relevant documents from large datasets. Despite proving useful in a variety of settings, these works suffer from the following limitations: (i) The documents are embedded in a non- interpretable space. (ii) There is no clear visually intuitive indication of how similar two retrieved documents are, i.e., black box retrieval. (iii) Vi- sualizing the embeddings requires dimensionality reduction techniques (Hotelling, 1933; Maaten and Hinton, 2008). By contrast, we propose a method that embeds medical text into a universal, small dimensional …

Content truncated to 500 words