Keywords:
Radioprotection / Radiation dose, Thorax, Respiratory system, CAD, CT, Computer Applications-Detection, diagnosis, Dosimetry, Cancer, Multidisciplinary cancer care
Authors:
A. Christe1, L. Leidolt1, A. Huber1, P. Steiger1, Z. Szucs-Farkas2, J. Roos3, J. Heverhagen1, L. Ebner1; 1Bern/CH, 2Biel/ Bienne/CH, 3Winterthur/CH
DOI:
10.1594/ecr2013/C-2614
Results
Individual nodule detection (first reader):
Individual standard CT sensitivities for lung nodules were 92.2%,
90.3%,
82.0%,
83.3% and 67.6% for reader 1,
2,
CAD1,
CAD2,
CAD3,
respectively (Table 1).
Only the sensitivity of CAD3 is significantly lower than radiologists’ sensitivity (p<0.001).
Sensitivity at lowest dose level decreases to 87.8%,
82.4%,
76.6%,
64.1%,
and 67.8% for reader 1,
2,
CAD1,
CAD2,
CAD3,
respectively (Fig.
1). Only CAD2 showed a significant loss of sensitivity at lowest dose level (64.1%,
p-value=0.028).
Combined nodule detection (first and second reader)
Combining reader 1 and 2 sensitivity increased to 96.8%,
which was a significant rise for reader 2 (p=0.016) but not for reader 1 (p=0.063).
This combined sensitivity dropped insignificantly to 94.0% at the lowest dose level.
Highest sensitivities at standard dose level - between 96.6% and 99.0% - were achieved by combining any radiologist with any CAD with a significant increased detection rate for both sides (p<0.05).
This combination worked at all dose levels with increased sensitivities for both sides (Table 1).
No significant loss of detectability with this combination was found at lower dose levels compared to standard CT.
Combining any two CADs,
sensitivities were significantly lower (between 83.3% and 89.4% for standard CT and between 75.6% and 86.4% for 25 ref mAs/80 kVp; p<0.001).
Sensitivity for solid nodules was around 90% for both radiologists and CADs,
but CADs detected significantly less GGN compared to radiologists: At standard CT sensitivity for GGN was 96.1%,
88.3%,
46.2%,
59.1% and 8.0% for reader 1,
2,
CAD1,
CAD2,
CAD3,
respectively.
Inter-observer agreement
Inter-observer agreement of the two radiologists was excellent with a Kappa level of 0.91±0.05 (standard error) at standard CT (Table 2).
Kappa strength of agreement did not drop significantly at lower dose levels (Kappa: 0.8 to 0.93).
Inter-observer agreement was lower between radiologists and CADs.
The mean Kappa strength of agreement between radiologists and CADs was 0.58 ±0.11 (p<0.0001) at standard CT.
This agreement did not change significantly for the other dose levels.
For all dose levels combined CAD3 scored the lowest agreements with the radiologists (0.45±0.1); CAD1 and CAD2 demonstrated an agreement of 0.67±0.09 and 0.59±0.11,
respectively.