FungAI: Leveraging artificial intelligence to identify fungi strains

Share this article

Did you know that there are over 100,000 species of fungi in the world? From the mushrooms we eat, to those that cannot be seen with the naked eye, identifying them is no easy feat. Yet it is crucial to identify the species accurately given that there are over 4,000 clinically important species that may cause life-threatening diseases.

At the forefront of Artificial Intelligence (AI) advancement in healthcare, Singapore General Hospital’s (SGH) Department of Microbiology and Synapxe’s Data Analytics and AI department formed a project team to collaborate on the development of a Fungal Species Detection using Artificial Intelligence (FungAI), a promising computer vision AI model being used in trails to medically identify common fungal species from patients’ specimens.

The project team from SGH and Synapxe (L to R): Ms Tan Mei Gie (SGH), Mr Benny Yip (SGH), Ms Millie Goh (Synapxe), Ms Christine Ang (Synapxe), Mr Benedict Lim (Collaborator), Mr Ho Yew Kay (Synapxe), Dr Tan Yen Ee (SGH), Mr Chia Kuok Wei (SGH), Dr Jeremy Ng (SGH), Dr Chan Chee Seng (SGH), Mr Lawrence Lin (SGH), Ms Samantha Tan (SGH), Dr Goh Han Leong (Synapxe), Prof Koh Tse Hsien (SGH) and Prof. Tony Lim (SGH).

Currently in pilot phase, FungAI aims to digitalise mycology identification processes and enhance diagnostic capabilities through AI technology. This will boost the overall efficiency of mycology diagnostics within the hospital, with the promise to improve diagnostic accuracy, streamline laboratory workflows and enhance laboratory capabilities.

Unlocking the potential of FungAI

FungAI aims to speed up diagnosis and improve laboratory productivity by maximising laboratory resources and efficiencies using AI.

Here are three ways FungAI can potentially improve healthcare delivery:

1. Accelerate fungal identification for commonly encountered moulds with the potential to reduce turnaround time for diagnostic results.

2. Empower laboratory staff to confidently perform basic mycology, ensuring timely and accurate results for patients and healthcare professionals.

3. Streamline laboratory workflows and enhance productivity gain for common fungal identification through digitalisation.

“With the rise in antifungal resistance, it is important to identify fungal species at a faster rate for earlier appropriate intervention. The FungAI technology aims to empower laboratories with limited resources by reducing the reliance on skilled workers which may take many years of training and experience to be proficient in the specialty. This is especially important given the difficulties of getting trained staff in both the current and future job market.”

- Dr Tan Yen Ee, Senior Consultant, Department of Microbiology, Singapore General Hospital

Innovating fungi identification

The procedure to identify fungal species is lengthy and heavily reliant on manual processes. The project team tested the concept with a particular fungal strain with a turnaround time of about four days and co-designed a solution that could halve the time required, allowing for earlier diagnosis and potential productivity gains.

In October 2019, SGH and Synapxe began experimenting with over 8,500 images collected over two years involving five species of commonly encountered fungal species. Through the experiment, it demonstrated that the turnaround time for identification could be reduced to just two days, with an accuracy of 80-90%.

The team endeavours to scale the experiment into a proof-of-concept, focusing on a broader range of fungal species, an efficient data collection process and a simple user interface.

If successful, FungAI may be deployed in the SGH mycology laboratory to address the shortage of skilled expertise in this area by allowing junior laboratory staff to identify commonly encountered fungal species by simply uploading a microscope image into the AI model and easing trained staff for complicated fungal cases. As the pilot project progresses, more details will be shared at a later stage.

“While most AI efforts in this area have leaned on established methods like Deep Learning Convolution Neural Network (CNN), our study explores the potential of Vision Transformer (ViT), a newer and promising approach. It's like adding a new tool to the healthcare toolbox. We found that combining the strengths of both CNN and ViT through an advanced machine learning technique, it enhances the accuracy of fungal classification, paving the way for more effective treatments”

- Dr Goh Han Leong, Senior Principal Specialist, Platform Services (Data Analytics & AI), Synapxe