Envisaging early detection for childhood retinoblastoma
- Wits News
Wits research team initiates a study exploring the use of AI for early detection of eye cancer.
Retinoblastoma is an eye cancer that is easily curable in the early stages, but sometimes deadly in South Africa, due to late detection.
Researchers and clinicians who work with childhood cancer are constantly pushing through barriers to deliver better chances of survival among child cancer patients. In one such initiative, researchers are investigating whether AI has the potential to support early detection of retinoblastoma, a type of cancer that begins in the retina and mostly affects children under the age of three. It accounts for about 5% to 6% of childhood cancers, amounting to roughly 50 to 100 retinoblastoma cases in South Africa each year.
Professor Jennifer Geel, the Academic Head of the Division of Paediatric Haematology and Oncology at Wits University, a paediatric oncologist at the Wits Donald Gordon Medical Centre, and president of the International Society of Paediatric Oncology (SIOP) in Africa, is overseeing a new study that will investigate the use of AI in early detection in primary healthcare settings.
“Retinoblastoma is a cancer of the eye that is almost 100% curable, but in our setting, we are achieving only about 60% survival,” Geel says.
She explains that this lower survival rate is due to various factors including a lack of knowledge or misdiagnosis at primary healthcare facilities. Or, if the diagnosis is correct, the family is told that the child may lose an eye, and fear then prevents them from seeking out further medical care, even though it would be life-saving.
“We want children to reach us quickly so that we can counsel the family properly, explain the options and show them that a child can still have an excellent outcome,” says Geel. “A good prosthesis can look so natural that people do not even notice it.”
Expediting the diagnosis with AI
There are many stories of retinoblastoma first being identified in children by accident. In response to flash photography, it can produce a white or pink glow in the pupil of the eye, known as leukocoria, which is then often investigated further by concerned parents. Other symptoms include crossed or misaligned eyes (strabismus), redness or swelling, or a protruding eye (buphthalmos). Because these primary symptoms are visible, they can potentially be detected and documented without advanced equipment.
Unfortunately, delivering education and training in detection at a national or continental level is a massive and complex task, but AI could offer a way to address this challenge. Geel is supervising Dr Lara Sandri, MD, a PhD candidate, who is studying the use of AI in detecting retinoblastoma. The process involves taking a simple smartphone photograph of the eye and running it through the AI detection system. If the system identifies a possible retinoblastoma, the child can promptly be referred to the relevant hospital or specialist.
At present, affected families experience multi-level institutional delays, even when they do commit to seeking help for their child. “A family may go from a GP to a clinic, then to a secondary or district hospital, then to ophthalmology, and eventually to us,” Geel says. In future, she explains, the tool should be able to identify a possible or likely case of retinoblastoma, and the child could then be sent directly to the correct specialist team.
The smartphone-based AI tool would ultimately be user-friendly enough to be used by primary healthcare nurses, clinic workers or optometrists, without the need for specialist oncologist supervision. This matters, Geel explains, because there are few formally internationally fellowship trained ocular oncologists in Africa. Even ophthalmologists are scarce here, with only 2.5 ophthalmologists per million people in sub-Saharan Africa, compared to 57 per million in the U.S.
“We are trying to expand that expertise and make it more widely available,” says Sandri. “AI may help us to do more with less.”
A 33-country survey co-authored by Geel, Sandri and Dr Jacques van Heerden evaluated AI readiness among African healthcare workers across the continent. It revealed that while 76% expressed strong willingness to adopt AI tools, 80% cited infrastructure deficits and 87% noted cost constrains as major barriers. This highlights the broader need for local capacity-building beyond just the development of the AI detection tool. But the tool could be an important step forward with a relatively minimal investment in technology.
Some years away from delivery
The study holds great promise for delivering life-saving early detection in varied healthcare settings, but it is still in its first phase of determining whether the AI tool can detect retinoblastoma, with the first patient having been enrolled in the study.
The next phase will assess whether the AI model can determine the stage of the tumour – whether it is in group A, B, C, D or E. Geel explains that group A is highly salvageable, while at group E, the eye generally has to be removed.
This long-term vision in addressing challenges at a grassroots level is essential to the development of solutions and the advancement of healthcare in every community. While the ongoing engagement of child patients with cancer and their families could become overwhelming for specialists, Geel is buoyed by the knowledge that she is making a tangible difference.
“To me it is almost a privilege. Not everyone can do this work. You have to be tough enough to put your emotions aside when you need to make clinical decisions. If emotion leads the decision, you may not make the right one. At the same time, you have to hold everything in mind: the cost, the psychosocial circumstances, the hospital, the family and the child. There is a great deal going on at once and you have to be clear-headed.”