Methods for automatic generation of radiological reports of chest radiographs: a comprehensive survey

Generation of a clear, correct, concise, complete, and coherent linguistic description of the visual patterns in a medical image is a challenging task. Unfortunately, many radiologists fail to satisfactorily perform this task due to various reasons such as workload, scant time, and fatigue. Although AI-based computer-aided detection (CADe) and computer-aided diagnosis (CADx) systems have been developed for observing and interpreting patterns in medical images, they do not generate radiological reports. In recent years, a lot of research has been done to develop automated report generation methods. This paper presents a comprehensive survey of all such methods specifically developed for chest radiographs. It consolidates information about standard chest X-ray datasets, state-of-the-art report generation methods, evaluation metrics, and their results. Deep learning-based techniques for automatically generating chest radiographic reports have been classified and discussed in detail. The encoder-decoder-based techniques have been meticulously categorized for a better understanding of the developments in this area. This paper is also beneficial for the researchers interested in developing automatic report generation systems for imaging modalities other than chest radiographs.

Automated summary

Generating accurate and coherent linguistic descriptions of visual patterns in medical images is a challenging task, as many radiologists struggle due to workload, time constraints, and fatigue. Research has been conducted in recent years to develop methods for automated report generation to tackle this issue.