Schematic block diagram of the proposed CAD system.
FR algorithm: Positive- and negative-affected points by gradient at point : Radius .
MIP processing of a stack of images along direction. Maximum value encountered along the axis is assigned to the final image.
Effects of MIP processing on nodules images. Upper row shows a true positive, while lower row shows a false positive; raw images on the left, while MIP images on the right.
Feature extraction: MIP generation from 3D volume and extraction of Zernike moments from MIP images.
Distribution of diameters of the annotated nodules in the studied LIDC database subset.
FROC curve for the overall CAD performance.
ANODE09 FROC curves for the presented CAD system.
Distribution of nodule characteristics for AL4. Values in the first column are rounded means of the evaluations given by radiologists for a given nodule. Characteristic value meanings are shown in Table II.
Value meanings for LIDC characteristics.
CAD system results for the detection part and heuristic FPR at AL 4. Sensitivity and FPs/patient are reported for each module of the CAD.
FROC curve analysis. Sensitivity at various FP/patient rate. Uncertainty estimations are based on binomial proportion using Wilson score interval (Ref. 59) with correction for continuity, with 95% confidence.
Sensitivity distribution for different types of nodule characteristics for AL4, in the working point with 71% overall sensitivity. Values in the first column are rounded means of the evaluations given by radiologists for a given nodule. Uncertainty estimations are based on binomial proportion using Wilson score interval (Ref. 59) with correction for continuity, with 95% confidence. Characteristic value meanings are shown in Table II.
Performance comparison with other studies using the LIDC dataset. Sensitivity distribution for different ALs, in the working point with 4 FP/patient, unless differently reported. Uncertainty estimates are shown where available. Reference 50 does not include ground glass opacities (GGOs).
Comparison of scores in ANODE09 study. Assigned scores are calculated as the average of true positive fractions at 1/8, 1/4, 1/2, 1, 2, 4, and 8 FP/patient. This work is referred to as “MIG GROUP.” Compared CAD systems are described in Ref. 18 apart from “Flyer Scan,” which is described in Ref. 57. In this table, nodule types use ANODE09 nomenclature. In particular, juxta-vascular and juxta-pleural nodules are described here as vascular and pleural nodules, respectively.
Article metrics loading...
Full text loading...