Table 1 Dice scores evaluating the performance of the standard and pathology-aware methods across different test sets (Healthy and Progstar). The results are derived using three models (UNet, DeepLabv3, and ReLayNet) and are reported for individual retinal layers-Inner Retina (IR), Outer Nuclear Layer (ONL), Photoreceptor Inner Segment (PR-IS), Photoreceptor Outer Segment (PR-OS), and Retinal Pigment Epithelium (RPE)-as well as an overall score. Each score includes mean and standard deviation values.

From: Looking outside the box with a pathology aware AI approach for analyzing OCT retinal images in Stargardt disease

Method

Test set

Model

Dice score

IR

ONL

PR-IS

PR-OS

RPE

Overall

Standard

Healthy

UNet

0.98±0.01

0.96±0.02

0.95±0.02

0.94±0.02

0.91±0.02

0.95±0.01

DeepLabv3

0.98±0.01

0.96±0.02

0.94±0.02

0.94±0.02

0.91±0.02

0.95±0.01

ReLayNet

0.98±0.01

0.96±0.02

0.95±0.02

0.94±0.01

0.91±0.02

0.95±0.01

Progstar

UNet

0.86±0.31

0.80±0.28

0.79±0.20

0.80±0.23

0.81±0.28

0.81±0.21

DeepLabv3

0.78±0.38

0.73±0.35

0.67±0.32

0.69±0.33

0.72±0.35

0.72±0.34

ReLayNet

0.81±0.35

0.82±0.25

0.73±0.28

0.77±0.27

0.81±0.29

0.79±0.24

Pathology-aware

Healthy

UNet

0.98±0.01

0.96±0.02

0.95±0.02

0.94±0.01

0.91±0.02

0.95±0.01

DeepLabv3

0.98±0.01

0.96±0.02

0.94±0.02

0.94±0.01

0.91±0.02

0.95±0.01

ReLayNet

0.98±0.01

0.96±0.02

0.94±0.02

0.94±0.02

0.91±0.02

0.95±0.01

Progstar

UNet

0.97±0.10

0.94±0.11

0.90±0.13

0.91±0.11

0.92±0.10

0.93±0.10

DeepLabv3

0.92±0.23

0.91±0.19

0.89±0.16

0.89±0.16

0.89±0.18

0.90±0.15

ReLayNet

0.98±0.07

0.94±0.10

0.91±0.11

0.91±0.09

0.92±0.11

0.93±0.09