Table 5.
Performance results of the models trained to discriminate between 35 and 43–47. At the top, we have the rank and performance of the models associated with the top 10 features, measured after training using only one feature. In the middle and bottom, we see performance metrics of the models measured after training with one to five features selected using the forward feature selection approach. The middle section corresponds to when only static metrics were used, while the bottom section includes both static and dynamic metrics. For all the evaluations, a repeated random sub-sampling validation approach was employed (50 repetitions). Each task/dataset had a testing dataset with 30% of the data, ensuring a minimum of 10 eyes of each class in the testing set.
| Individual Features Performance | ||||||
| Rank | Feature | AUC | bACC | Sensitivity | Specificity | |
| 1 | MAs | 0.8 ± 0.06 | 0.75 ± 0.07 | 0.7 ± 0.14 | 0.79 ± 0.1 | |
| 2 | MAT_6_months | 0.8 ± 0.06 | 0.7 ± 0.07 | 0.68 ± 0.11 | 0.72 ± 0.1 | |
| 3 | MAT_12_months | 0.79 ± 0.07 | 0.75 ± 0.06 | 0.66 ± 0.12 | 0.85 ± 0.09 | |
| 4 | //_FAZCircularity | 0.79 ± 0.07 | 0.71 ± 0.07 | 0.75 ± 0.11 | 0.67 ± 0.14 | |
| 5 | //_PD_InR_DCP | 0.72 ± 0.07 | 0.67 ± 0.07 | 0.67 ± 0.12 | 0.67 ± 0.13 | |
| 6 | //_÷[SCP_DCP]_PD_InR | 0.71 ± 0.08 | 0.66 ± 0.08 | 0.64 ± 0.12 | 0.67 ± 0.12 | |
| 7 | //_FAZPerimeter | 0.69 ± 0.12 | 0.62 ± 0.1 | 0.69 ± 0.16 | 0.55 ± 0.19 | |
| 8 | //_ VD_InR_DCP | 0.68 ± 0.08 | 0.64 ± 0.06 | 0.66 ± 0.11 | 0.61 ± 0.12 | |
| 9 | ÷[VD_PD]_InR_DCP | 0.65 ± 0.1 | 0.6 ± 0.1 | 0.56 ± 0.15 | 0.64 ± 0.15 | |
| 10 | //_÷[SCP_DCP]_VD_InR | 0.64 ± 0.11 | 0.61 ± 0.09 | 0.62 ± 0.14 | 0.61 ± 0.15 | |
| FFS Performance - Static features | ||||||
| #features | AUC | bACC | Sensitivity | Specificity | Features | |
| 1 | 0.8 ± 0.06 | 0.75 ± 0.07 | 0.7 ± 0.14 | 0.79 ± 0.1 | MAs | |
| 2 | 0.84 ± 0.07 | 0.79 ± 0.07 | 0.76 ± 0.14 | 0.82 ± 0.1 | + PD_InR_SCP_6 × 6 | |
| 3 | 0.84 ± 0.07 | 0.81 ± 0.08 | 0.79 ± 0.13 | 0.83 ± 0.12 | + triglycerides | |
| 4 | 0.84 ± 0.07 | 0.81 ± 0.08 | 0.8 ± 0.14 | 0.83 ± 0.11 | + ÷[VD_PD]_OutR_SCP_6 × 6 | |
| 5 | 0.84 ± 0.07 | 0.82 ± 0.08 | 0.81 ± 0.13 | 0.83 ± 0.12 | + RT_Inr_GCL | |
| FFS Performance - Static + Dynamic features | ||||||
| #features | AUC | bACC | Sensitivity | Specificity | Features | |
| 1 | 0.8 ± 0.06 | 0.75 ± 0.07 | 0.7 ± 0.14 | 0.79 ± 0.1 | MAs | |
| 2 | 0.87 ± 0.05 | 0.78 ± 0.06 | 0.76 ± 0.12 | 0.79 ± 0.1 | + //_FAZCircularity | |
| 3 | 0.89 ± 0.06 | 0.83 ± 0.06 | 0.83 ± 0.09 | 0.82 ± 0.11 | + //_ PD_InR_DCP_6 × 6 | |
| 4 | 0.91 ± 0.05 | 0.81 ± 0.09 | 0.81 ± 0.14 | 0.8 ± 0.12 | + ÷[SCP_DCP]_PD_OurR_6 × 6 | |
| 5 | 0.91 ± 0.05 | 0.84 ± 0.08 | 0.84 ± 0.12 | 0.83 ± 0.11 | + //_÷[VD_PD]_InR_SCP_6 × 6 | |
Notation: //—slope measured between the baseline and the 12-month follow-up; MA—microaneurysm; MAT—microaneurysm turnover; VD—skeletonized vessel density; PD—perfusion density; SCP—superficial capillary plexus; DCP—deep capillary plexus; InR—inner ring; OutR—outer ring; 6 × 6—6 × 6 mm angiography acquisition protocol; ÷[SCP_DCP]—ratio between the values measured at the SCP and DCP; ÷[VD_PD]—ratio between the measured values of VD and PD; FAZCircularity—circularity of the foveal avascular zone; FAZPerimeter—perimeter of the foveal avascular zone.