19 June 2020
Researchers from Monash University, Sun Yat-sen University and The First Affiliated Hospital of Harbin Medical University have developed an advanced medical report generation method using AI technology. This will accurately read lung CT images and help diagnose COVID-19 cases.
In addition to the Reverse Transcription Polymerase Chain Reaction (RT-PCR) tests, lung CT scan analysis is another essential testing method for COVID-19 – which has infected more than seven million people globally. A key step in controlling the infection is to quickly identify positive cases and treat patients appropriately.
“While the number of confirmed COVID-19 cases continues to outweigh the number of experienced medical personnel needed to diagnose lung CT scans, an automated report generation tool allows for more accurate and timely COVID-19 diagnoses,” explains Senior Research Fellow Dr Xiaojun Chang from the Faculty of Information Technology (IT), Monash University.
Image captioning is a form of AI technology which summarises visual information, such as images or videos, in an automatically generated description. This complex task requires the AI model to bridge visual and linguistic information to deliver accurate analyses.
In medical report generation, the AI model requires an increased capacity to understand the medical domain and describe medical images in a coherent way, covering appropriate terminologies.
“Our research has formulated the first public COVID-19 CT report dataset, and proposed a novel deep learning model which demonstrates how an AI-generated method is capable of producing accurate and robust medical reports,” explains PhD student Mingjie Li from Monash University’s Faculty of IT.
Rather than simply notifying a patient that they may be diagnosed with COVID-19, a healthcare professional can use the AI-generated report to offer a complete picture of what the diagnosis might look like – taking into account additional factors such as fever, coughing or breathing difficulties, which are not identified in a medical image alone.
The comprehensive reports generated by this AI method enhances the credibility of the doctor's diagnosis and provides patients with accurate information relating to their health.
“This research has wider clinical value as well as future applications across other diseases. While an AI-generated medical report will never eliminate the work of a healthcare professional, it can act as an effective tool to reduce the pressure experienced by doctors during COVID-19,” said Professor Anyuan Li, a radiologist from The First Affiliated Hospital of Harbin Medical University.
Findings of this research have been submitted to the thirty-fourth Annual Conference on Neural Information Processing Systems in December.
To view the research paper, please visit: https://arxiv.org/abs/2006.03744