CSIRO research identifies AI models to improve automated chest X-ray diagnostics

in 2023 November 20
News release

New research from Australia’s national science agency CSIRO has identified methods to improve artificial intelligence (AI) diagnosis of heart and lung disease using X-rays.

A a recent paperCSIRO Australia e. Health Research Center (AEHRC) researchers compared different AI models to better understand and determine the diagnostic accuracy of automated chest X-ray interpretation and reporting.

CSIRO research scientist and lead author of the paper, Dr. Aaron Nicolson said that a better understanding of optimal models will lead to more accurate use of AI to diagnose X-ray images.

“AI can improve health services and especially better support healthcare professionals, reducing the burden and workload of their current unautomated practice,” said Dr.

“Automated reporting for X-rays can reduce physician burnout and give them room to provide more reliable patient care.” The study shows future potential to better support physicians.

Current methods of generating AI X-ray reports use an “encoder” to read chest X-ray images and a “decoder” to produce the report. To date, no studies have been conducted on which encoding and decoding devices are best suited for the automatic generation of chest X-ray reports.

In addition, knowledge gained from a single task, such as classifying natural images or generating Wikipedia articles, can be transferred to improve the task at hand. In this case, automated reporting. This method is known as “warm-starting” the AI ​​model.

In a world first, the CSIRO AEHRC imaging team tested different encoders and decoders, as well as the performance of different tasks, in a warm start to the chest x-ray report generation task.

The findings show that the optimal combination of encoder and decoder together with the warm-start method allows 26.9 percent. relatively improve the accuracy of automatic image reporting. The assessment was made in comparison with the findings of a human radiologist.

Radiologist Dr Doug Anderson of Monash Medicine, Victoria, said: “Clinical burnout is a risk factor for mental illness and is particularly prevalent in radiologists due to their high workload and demanding clinical documentation.

“Increasing clinical reliance on imaging for diagnosis and the relative shortage of radiologists is creating an unsustainable workload and a search for workload management solutions.”

“An exciting potential solution to the complex workload of a radiologist is the use of artificial intelligence to assist in the interpretation of chest X-rays and documentation.”

Although the model consistently identifies some pathologies (e.g., pleural effusion), others (e.g., lung injury) are not yet accurately identified.

The next step for the researchers is to improve the AI ​​model so that it can accurately identify most pathologies. These improvements are needed before the technology can be used in a clinical setting.

Godfrey Kemp

"Bacon fanatic. Social media enthusiast. Music practitioner. Internet scholar. Incurable travel advocate. Wannabe web junkie. Coffeeaholic. Alcohol fanatic."