Table 3. A comparison of the response classification results obtained using tumor size alone (RECIST criteria), using KNN-based QUS feature combination, and using KNN-based QUS feature combinations with the addition of pretreatment data. Reported are sensitivity (Sen), specificity (Spe), and accuracy (Acc) mean ± standard deviation.
| Pre-Tx | Week 1 | Week 4 | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Sen | Spe | Acc | Sen | Spe | Acc | Sen | Spe | Acc | |
| RECIST | NA | 16 | 60 | 30 | 53 | 50 | 52 | ||
| ΔQUS | NA | 61 ± 13 | 59 ± 9 | 60 ± 10 | 79 ± 10 | 76 ± 11 | 77 ± 8 | ||
| ΔQUS + QUSw0 | 67 ± 13 | 63 ± 7 | 65 ± 9 | 76 ± 11 | 64 ± 11 | 70 ± 9 | 80 ± 9 | 79 ± 5 | 80 ± 5 |
| p-value | NA | 0.03* | 0.33 | ||||||
The results (sensitivity, specificity, and accuracy in percentages) are reported for weeks 1, 4 and 8 obtained from leave-one-out analysis cross-validation. ΔQUS represents [ΔMBF ΔSS ΔSAS] and QUSw0 represents [MBFw0 SSw0 SASw0]. The last row presents the p-value significance of the difference between the mean accuracies of ΔQUS and ΔQUS + QUSw0 KNN models. Reported values are mean and standard deviation of the accuracies obtained by running the classification 10 times using 10 bootstrap samples from responder group.