# That's the Wrong Lung! Evaluating and Improving the Interpretability of Unsupervised Multimodal Encoders for Medical Data

arXiv: https://arxiv.org/abs/2210.06565 · HTML: https://mlsanity.com/paper/2210.06565/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 33 references that are in the map.
- First posted on arXiv 12 Oct 2022.

## Builds on (most cited first, 22 of 22)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Neural Machine Translation by Jointly Learning to Align and Translate](https://mlsanity.com/paper/1409.0473/) (2014)
- [LXMERT: Learning Cross-Modality Encoder Representations from Transformers](https://mlsanity.com/paper/1908.07490/) (2019)
- [VisualBERT: A Simple and Performant Baseline for Vision and Language](https://mlsanity.com/paper/1908.03557/) (2019)
- [Publicly Available Clinical BERT Embeddings](https://mlsanity.com/paper/1904.03323/) (2019)
- [VL-BERT: Pre-training of Generic Visual-Linguistic Representations](https://mlsanity.com/paper/1908.08530/) (2019)
- [Attention is not Explanation](https://mlsanity.com/paper/1902.10186/) (2019)
- [Attention is not not Explanation](https://mlsanity.com/paper/1908.04626/) (2019)
- [MIMIC-CXR-JPG, a large publicly available database of labeled chest radiographs](https://mlsanity.com/paper/1901.07042/) (2019)
- [Is Attention Interpretable?](https://mlsanity.com/paper/1906.03731/) (2019)
- [Confounding variables can degrade generalization performance of radiological deep learning models](https://mlsanity.com/paper/1807.00431/) (2018)
- [Large-Scale Adversarial Training for Vision-and-Language Representation Learning](https://mlsanity.com/paper/2006.06195/) (2020)
- [Pixel-BERT: Aligning Image Pixels with Text by Deep Multi-Modal Transformers](https://mlsanity.com/paper/2004.00849/) (2020)
- [ScispaCy: Fast and Robust Models for Biomedical Natural Language Processing](https://mlsanity.com/paper/1902.07669/) (2019)
- [TieNet: Text-Image Embedding Network for Common Thorax Disease Classification and Reporting in Chest X-rays](https://mlsanity.com/paper/1801.04334/) (2018)
- [Joint Modeling of Chest Radiographs and Radiology Reports for Pulmonary Edema Assessment](https://mlsanity.com/paper/2008.09884/) (2020)
- [A Comparison of Pre-trained Vision-and-Language Models for Multimodal Representation Learning across Medical Images and Reports](https://mlsanity.com/paper/2009.01523/) (2020)
- [Multimodal Representation Learning via Maximization of Local Mutual Information](https://mlsanity.com/paper/2103.04537/) (2021)
- [Multimodal Integration of Human-Like Attention in Visual Question Answering](https://mlsanity.com/paper/2109.13139/) (2021)
- [Improving Joint Learning of Chest X-Ray and Radiology Report by Word Region Alignment](https://mlsanity.com/paper/2109.01949/) (2021)
- [Weak Supervision helps Emergence of Word-Object Alignment and improves Vision-Language Tasks](https://mlsanity.com/paper/1912.03063/) (2019)

## Abstract (arXiv)

Pretraining multimodal models on Electronic Health Records (EHRs) provides a means of learning representations that can transfer to downstream tasks with minimal supervision. Recent multimodal models induce soft local alignments between image regions and sentences. This is of particular interest in the medical domain, where alignments might highlight regions in an image relevant to specific phenomena described in free-text. While past work has suggested that attention "heatmaps" can be interpreted in this manner, there has been little evaluation of such alignments. We compare alignments from a state-of-the-art multimodal (image and text) model for EHR with human annotations that link image regions to sentences. Our main finding is that the text has an often weak or unintuitive influence on attention; alignments do not consistently reflect basic anatomical information. Moreover, synthetic modifications -- such as substituting "left" for "right" -- do not substantially influence highlights. Simple techniques such as allowing the model to opt out of attending to the image and few-shot finetuning show promise in terms of their ability to improve alignments with very little or no supervision. We make our code and checkpoints open-source.