Artificially intelligent detection of retinal pigment sign using P3S-Net for retinitis pigmentosa analysis

• P3S-Net can identify the retinal pigment signs without preprocessing in the fundus images. • It is a multiscale segmentation network based multiple streams, which utilizes only 155 K trainable parameters. • The four skip connections transfer a low-level feature, which enables the network to identify PS that are small. • The proposed P3S-Net was trained from scratch without using any weight initialization or migration for transfer learning. Image segmentation is a major and crucial problem in medical imaging. It is the foundation for clinical diagnostic methods, treatments, and computer-assisted disease diagnosis (CAD). An ocular condition known as retinal pigmentosa (RP) first results in night blindness and persistent degradation of the retinal pigment. An advanced deep learning method for computer-aided disease diagnosis that detects pigment signs (PS) in RP patients, enabling ophthalmologists to identify and treat the condition on or before schedule. Many researchers now use costly CAD approaches; however, the suggested solution, which uses a fundus image dataset, is quick, easy, and affordable. This paper presents a large kernel pigment sign semantic segmentation network (P3S-Net) to enhance segmentation performance and provide a useful receptive field. P3S-Net is a residual network with less trainable parameters due to its large kernel architecture. Moreover, these large-kernel receptive fields yield improved segmentation results with early, intermediate, and late feature information fusion. P3S-Net has an order of magnitude less trainable parameters of 155k in certain scenarios and is smaller than current medical image segmentation techniques. Utilizing the publicly available Retinal Images for Pigment Signs (RIPS) dataset for retinal pigment detection and segmentation, 4-fold cross-validation tests were conducted to assess the suggested P3S-Net. The performance of the suggested network is frequently better and more competitive when compared to cutting-edge techniques.