As the world becomes increasingly technological, researchers and scientists are leveraging these advances for scientific improvement. Artificial intelligence (AI) is now being used to develop products, screen drug candidates, or monitor patients. One subset of AI called deep learning has been particularly impactful. Amazon Web Services (AWS) explains that deep learning is an AI network that trains computers to process in a way that is modeled after the human brain with many artificial neural networks that process and analyze data.
Utilizing Deep Learning for RP Monitoring
An article in Healio Ophthalmology shares that deep learning could also be used to objectively monitor disease progression in conditions like retinitis pigmentosa (RP), a group of inherited diseases characterized by retinal degeneration. Normally, the retina converts light into electric signals; your brain then reads this as vision. When photoreceptor cells in the retina degenerate, vision is lost. Most people with RP are considered legally blind by the time they are 40 years old. They often struggle with visual acuity, color perception, peripheral vision, night vision, and central vision.
Research is looking into potential interventions. However, RP is largely managed by monitoring progression and attempting to preserve visual function. This is where deep learning could help. A retrospective study in JAMA Ophthalmology discussed this use case. Researchers sourced ultra-widefield fundus autofluorescence (UWFAF) images, ultra-widefield pseudocolor (UPWC) images, and a mix of the two from 695 patients with RP (for a total of 1,274 eyes). Using these images, the research team trained 31 deep learning models to assess visual acuity (how sharp your vision is).
The deep learning models were able to successfully estimate patients’ visual acuity and how sensitive their retinas were. More interestingly, shared the research team, UWFAF images on their own were the best at enhancing accuracy. These images offer benefits for clinicians, who can gather them easily, and for patients, as getting the images requires no invasive procedures.
