Similarly to RETOUCH, the Pegasus system yielded a high performance for detection of PED, likely due to its distinct appearance and location in relation to the RPE. For this task, Pegasus relatively outperformed on SRF (particularly on data set B) and IRF (on data set A) but underperformed on IRF (on data sets B and C). Again, these are comparable to Pegasus’s average performance across the three data sets. For the fluid segmentation task, the mean Dice coefficients for Spectralis images were 0.69 for IRF, 0.57 for SRF, and 0.68 for PED. For the fluid detection task, the results were comparable for PED and SRF (on data set B), although Pegasus relatively outperformed on IRF (on data sets A and B) and underperformed on SRF (on data set A) and IRF (on data set C). 18 A comprehensive review of algorithm performance in these tasks for each of the three fluid subtypes, plus a table of other relevant works, can be found elsewhere. The RETOUCH project presents the performance of eight different deep learning algorithms for (a) automated fluid detection and (b) automated fluid segmentation. These images originated from standard spectral domain OCT devices, including the Heidelberg Spectralis (Heidelberg, Germany), Zeiss Cirrus (Dublin, California, USA), Nidek (Gamagori, Japan), and Topcon Maestro (Tokyo, Japan). Training was conducted using images that were graded by between one and five UK board-certified ophthalmologists providing the ground truth. The abnormal OCT scans included retinal pathologies such as AMD, DME, macular hole, and epiretinal membrane. Training was conducted using large heterogeneous real-world data sets from multiple clinical organizations across the world and incorporated both normal and abnormal scans. To minimize the effect of high-frequency image noise (speckle), areas of detected fluid were designated to be a minimum of 5 pixels in size. ![]() The models were trained using the DeeplabV3+ architecture, 13 allowing the system to identify different instances of the same class in the same image. This uses three convolutional neural network (CNN) models: one to predict SRF, one IRF, and one PED. Pegasus-OCT has automated fluid segmentation algorithms, aimed at delineating fluid within and beneath the retinal structures. The potential of Pegasus-OCT for automated fluid quantification and differentiation of IRF, SRF, and PED in OCT images has application to both clinical practice and research. Dice coefficients and sensitivity and specificity values indicate the potential for application to automated detection and monitoring of retinal diseases such as age-related macular degeneration and diabetic macular edema. The Pegasus automated fluid segmentation algorithms were able to detect IRF, SRF, and PED in SD-OCT b-scans acquired across multiple independent data sets. IRF detection in a third data set yielded less favorable agreement (0.46–0.57) and sensitivity (0.59–0.68), attributed to image quality and ground truth grader discordance. PED detection on the first data set showed moderate agreement (0.66) with high sensitivity (0.97) and specificity (0.98). Detection performance was assessed by calculating sensitivities and specificities, while Dice coefficients were used to assess agreement between the segmentation methods.įor two data sets, IRF detection yielded promising sensitivities (0.98 and 0.94, respectively) and specificities (1.00 and 0.98) but less consistent agreement with the ground truth (dice coefficients 0.81 and 0.59) likewise, SRF detection showed high sensitivity (0.86 and 0.98) and specificity (0.83 and 0.89) but less consistent agreement (0.59 and 0.78). Intraretinal fluid (IRF), sub-retinal fluid (SRF), and pigment epithelial detachment (PED) were automatically segmented by Pegasus-OCT for each b-scan where ground truth from data set owners was available. The Pegasus automated fluid segmentation algorithms were applied to three data sets with edematous pathology, comprising 750, 600, and 110 b-scans, respectively. ![]() ![]() To evaluate the performance of the Pegasus-OCT (Visulytix Ltd) multiclass automated fluid segmentation algorithms on independent spectral domain optical coherence tomography data sets.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |