Fundus image analysis of retinitis pigmentosa using artificial intelligence

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: InceptionV3 achieved the highest accuracy of 96.97% on the test set; VGG16 and ResNet50 performed comparatively but with lower accuracy (specific values not reported in abstract)

• Primary Outcome: Classification accuracy of CNN models for distinguishing RP from non-RP colour fundus images

• Nnt Or Sensitivity: Sensitivity and specificity not reported in abstract; accuracy 96.97% (InceptionV3); Grad-CAM attention maps showed clinically relevant focus on peripheral retina and posterior pole

• Confidence Interval: Not reported

Clinical Application

Colour fundus photography is widely available in Australian optometry practices, ophthalmology clinics, and diabetic retinopathy screening programs. The technical infrastructure for CNN-based image analysis exists. However, the model requires external validation on diverse Australian populations before clinical deployment. Regulatory approval from the TGA as a Software as a Medical Device (SaMD) would be required. RP affects approximately 1 in 4,000 Australians, with an estimated 15,000–20,000 individuals affected nationally. There is no PBS-listed pharmacological treatment for RP, though voretigene neparvovec (Luxturna) is TGA-approved for RPE65-associated RP and listed on the Life Saving Drugs Program. Early diagnosis is clinically important for genetic counselling, family screening, occupational advice, and eligibility for emerging gene therapies. RACGP guidelines do not currently include RP screening in routine health assessments. Australian optometrists and ophthalmologists routinely perform fundus photography, making the screening concept feasible in principle. The Macular Disease Foundation Australia and Retina Australia advocate for improved RP awareness and early detection. Any clinical deployment would require TGA SaMD registration, validation in Australian populations (including Aboriginal and Torres Strait Islander communities where consanguinity may increase RP prevalence), and integration with existing referral pathways to ophthalmology genetic services. Potentially applicable to adult patients undergoing routine colour fundus photography who may have undiagnosed RP; most relevant in optometry, general ophthalmology, and medical screening contexts. Current evidence is insufficient to support deployment beyond research settings.

Abstract

Retinitis pigmentosa (RP) is a group of inherited retinal diseases that are caused by genetic defects that lead to progressive photoreceptor loss and eventual blindness. Early diagnosis would be helpful for effective management of the disease; however, many patients stay unaware of early symptoms. Meanwhile, fundus images are widely obtained during routine medical checkups but are underused for detecting RP. This study explores the effectiveness of finetuning deep learning models, pre-trained for general visual tasks, to identify RP from color fundus images. The dataset comprised 321 color fundus images from 201 Japanese subjects at Keio University Hospital, including 200 images from 107 patients with retinitis pigmentosa and 121 images from 94 non-retinitis pigmentosa subjects. Multiple images were available for some subjects. Using transfer learning, pretrained convolutional neural network models -VGG16, Resnet50, and InceptionV3- were finetuned to detect RP. As a result, Inception V3 achieved the best accuracy of 96.97%, which matches the average diagnostic accuracy of ophthalmologists. Gradient-weighted Class Activation Mapping (Grad-CAM) suggested that the model attended to clinically relevant fundus regions, including the peripheral retina and posterior pole, which may reflect features such as peripheral degenerative changes and retinal vascular attenuation. These findings support the potential interpretability of the finetuned model and suggest that deep learning may assist ophthalmologists in RP screening as a supportive tool.

References

1. 1.Ubukata, S., Kanato, M., Katada, Y., Yang, L., Ozawa, N., Ibuki, M., Negishi, K., & Kurihara, T. (2026). Fundus image analysis of retinitis pigmentosa using artificial intelligence. PLOS ONE. https://doi.org/10.1371/journal.pone.0354452