Using AI to tackle invasive fungal infections

Dr Michelle Ananda-Rajah

by Anne Crawford

Detecting and managing invasive fungal disease (IFD) is a growing area of clinical importance in hospitals globally – particularly in an era of rising drug resistance – yet surveillance of these infections has proved challenging.

Researchers in the Department of Infectious Diseases, Central Clinical School (CCS), & General Medical Unit, Alfred Health, have used artificial intelligence (AI) to develop a fresh approach to what’s called anti-fungal stewardship.



The scientists, led by the CCS’s researcher/clinician Dr Michelle Ananda-Rajah, used a machine language-based technology called natural language processing (NLP) to detect and monitor invasive mold disease (IMD) in haematology patients at the Alfred Hospital. IMD, a difficult-to-detect fungal disease spread through spores, can cause life-threatening pneumonia in patients with weak immunity. It affects 10-15% of patients with acute leukaemia and is linked to a mortality rate of 30-80%.

Scan of lungs with fungal lesions obtained using NLP.
These data are now being used to train a deep learning image
classifier to better help clinicians pick up fungal infections.
Green circles are fungal lesions labeled by the radiologists.
Orange are NLP identified.

NLP using machine learning was programmed to ‘read’ chest CT (computed tomography) scan reports, which are performed when IMD is first suspected. More than 3000 CT reports from 1 September 2008 to 31 December 2017 were processed using NLP, identifying 205 IMD episodes in 185 patients, among them heavily immuno-compromised haemopoietic stem cell transplant recipients and blood cancer patients.

“Our overall aim of the study is to use NLP to detect fungal infections in real-time or as close to real-time as possible,” Dr Ananda-Rajah said. “We found that NLP is useful for population-level detection of fungal infection. It avoids restricting our focus on the highest risk groups only but includes everyone – challenging our preconceived notions of who is vulnerable to these infections.”

The researchers then looked at the positive results, identified haematology patients who had fungal infections and audited their care to look at areas that could be improved – across the whole hospital.

The method could help improve care, for example, by prompting clinicians to order a bronchoscopy earlier and faster than they might have done previously or by making sure a preventive antifungal medications are being administered correctly, said Dr Ananda-Rajah, who is a consultant physician in infectious diseases and general medicine at Alfred Health.

“Hospitals do not have a good means of monitoring these infections and understanding how they are being treated and what their outcomes are, despite spending millions of dollars on antifungal drugs for them,” she said.

“Machine learning could be working in the background finding patients with infections 24/7 and very efficiently,” she said.

“That’s something we’ve never had until now.”

The study, published last month in the Journal of Clinical Medicine, attracted considerable interest online. The method tested in it, known as FungalAI™, has been running in the Alfred Hospital since December 2018.

Dr Ananda-Rajah anticipates it will be adopted more broadly.

“I’m pretty proud of this work, it’s the culmination of many years of hard work.”

Dr Ananda-Rajah has been awarded a Monash Partners Medical Research Futures Fund (MRFF) TRIP Fellowship to trial the artificial intelligence platform technology; the study’s findings will now be validated in a real-world trial in seven major Australian hospitals, and the largest public hospital in Singapore, led by Monash University and The Alfred.

First author on this study was medical registrar Dr Diva Baggio. Associate Professor Reza Haffari from the Faculty of Information Technology, Monash University, Clayton, was pivotal in developing the NLP model used in the study.

Baggio D, Peel T, Peleg AY, Avery S, Prayaga M, Foo M, Haffari G, Liu M, Bergmeir C, Ananda-Rajah M. Closing the Gap in Surveillance and Audit of Invasive Mold Diseases for Antifungal Stewardship Using Machine Learning. J Clin Med. 2019 Sep 5;8(9). pii: E1390. doi: 10.3390/jcm8091390.