Table 2.
Using functional lung information to predict RILT
| Study first author | N | Age | Cancer type | FLI modality: Definition | RT modality | RILT predicted | Best prediction outcome using functional parameters | Model used to make best prediction | Improvement in prediction compared with nonfunctional information |
|---|---|---|---|---|---|---|---|---|---|
| Bin143 | 217 | 60.8 (mean) | 169 lung 48 esophageal |
CT V (HU): Weighting | N/A | Grade 2+ RP | AUC: 0.874 (95% CI, 0.871-0.877) | Dual-omics model, combining radiomics with deep learning | N/A |
| Dhami84 | 20 | 67.5 (med) | 15 NSCLC 2 SCLC 3 other |
SPECT Q: Thresholding from 5%-95% of max, in 5% increments | 3DCRT IMRT/VMAT PBT SBRT |
Grade 2+ RP | Sensitivity: 100% Specificity: 81.25% P = .008 |
Cutoff at 13.3 Gy, using perfused mean lung dose (pMLD) at the 70th percentile threshold | No (univariate) Yes (bivariate) |
| Ding73 | 40 | N/A | All NSCLC | SPECT Q: Thresholding from 10%-60% of max, in 10% increments Weighting | 3DCRT IMRT |
All grades RP | Thresholded AUC: 0.928 (95% CI, 0.842-1.013) Sensitivity: 90.9% Specificity: 86.2% Accuracy: 87.5% |
Cutoff at 20% of maximum, for fV20, was the most predictive | N/A |
| Dougherty86 | 31 | 64 (med) | 26 NSCLC 5 SCLC |
CT V (HU): 15% of max, with further processing | VMAT IMPT |
Grade 2+ RP Grade 3+ RP |
Grade 2+: IMPT: NTCP ↓ 5.7% VMAT: NTCP ↓ 6.2% Grade 3+: IMPT: NTCP ↓ 2.4% VMAT: NTCP ↓ 3.4% |
NTCP model from Faught et al89,90 | Yes |
| Farr152 | 71 | 67 (med) | All NSCLC | SPECT Q: 20%, 40%, 60%, and 80% of max | IMRT SBRT |
Grade 3+ RP | Baseline SPECT Q: AUC: 0.79 (95% CI, 0.68-0.91) Sensitivity: 72% Specificity: 70% Odds ratio: 7.8 Odds ratio with GTV: 9.2 Post-RT SPECT Q: AUC: 0.8 (95% CI, 0.62-0.94) |
Baseline: Perfusion defect score cutoff Post-RT: Difference in defect score cutoff |
N/A |
| Farr72 | 45 | 67 (median) | All NSCLC | SPECT Q: 20%, 40%, 60%, and 80% of max | IMRT SBRT |
Symptomatic RP | Spearman’s rs = 0.4, P = .02 Relative risk estimate: 3.6 (95% CI, 1.1-12) | Correlation analysis for rs is for perfusion reduction in 21-40 Gy bin. Relative risk estimate reflects all reductions. |
N/A |
| Farr121 | 58 | 67 (median) | All NSCLC | SPECT Q: Thresholding in 10% increments from 20th to 80th percentiles, followed by weighting | IMRT | Grade 2+ RP | Threshold odds ratio 1.53, P < .01 AUC = 0.81 (95% CI, 0.7-0.93) Weighting Odds ratio 1.4, P < .01 AUC = 0.78 (95% CI, 0.66-0.91) |
Best for OR both thresholded and weighted models was found when using fMLD. Best for AUC for both thresholded and weighted models was found when using fV30 |
Yes |
| Faught89 | 70 | N/A | All NSCLC | CT V (HU): 15% below average lung function | IMRT 3DCRT |
Grade 2+ RP Grade 3+ RP |
Grade 2+: AUC: 0.723, P < .01 NTCP ↓ 8%* (univariate) NTCP ↓ 10.4% (bivariate) Grade 3+: AUC: 0.674, P = .13 NTCP ↓ 4.8%* (univariate) NTCP ↓ 4.7% (bivariate) |
Grade 2+: Best AUC uses fMLD Best univariate NTCP reduction uses fV10 Best bivariate NTCP reduction uses fV10 and V10 Grade 3+: Best AUC uses fV20 Best univariate NTCP reduction uses fV30 Best bivariate NTCP reduction uses fV10 and V10 |
N/A |
| Faught90 | 70 | N/A | All NSCLC | CT V (HU): Thresholding from 5th to 95th percentile in increments of 5; nonlinear weighting | N/A | Grade 2+ RP Grade 3+ RP |
Grade 2+: AUC: 0.73, P < .01 (threshold) AUC: 0.74, P < .01 (weighting) Grade 3+: AUC: 0.7, P = .033 (threshold) AUC: 0.67, P < .03 (weighting) NTCP 10% reduction: 17.2% NTCP 20% reduction: 30.2% |
Grade 2+: Threshold best model used fV20 at 86th percentile or fMLD at 69th percentile. Sigmoid best model used amount of functional lung receiving ≥ 20 Gy (F20). Grade 3+: Threshold best model used fV20 at 85th percentile. Sigmoid best model used F20. Sigmoid NTCP 10% reduction requires a 17.2% reduction in F20 and a 20% reduction requires a 30.2% reduction in F20. |
Yes |
| Gayed174 | 50 | 67.6 (mean) | All lung cancer | Planar Q: Lung perfusion score | IMRT 3DCRT Proton |
Grade 2+ pulmonary complications (O2 dependence, respiratory failure, dyspnea, etc) | Lung perfusion score higher in patients with pulmonary complications, P = .01. Odds ratio = 1.6 (95% CI, 1.07-2.39) Odds ratio = 3.25 (95% CI, 1.37-7.70) (multivariate) AUC = 0.7 |
Lung perfusion score rates the perfusion defects seen in perfusion imaging | N/A |
| Hodge135 | 1 | N/A | NSCLC | He-MRI V: Automated threshold | IMRT | Grade 3+ RP | Predicted risk of grade 3+ RP was 4% with functional planning | NTCP model assessing damage to individual functional subunits179,180 | Yes |
| Hoover144 | 26 | N/A | 20 NSCLC 6 SCLC |
SPECT V/Q: Weighting | N/A | Grade 2+ RP | AUC: 0.74 (ventilation) (95% CI, 0.54-0.94) AUC: 0.74 (perfusion) (95% CI, 0.54-0.93) Significant correlations found with increase in functional parameters and RP |
N/A | Yes |
| Huang140 | 244 | N/A | All lung cancer | CT V (HU): Weighting | IMRT VMAT |
Grade 3+ RP | AUC: 0.77 Sensitivity: 0.71 Specificity: 0.76 |
Fully connected Convolutional neural network (CNN) | Yes |
| Huang136 | 36 | 66 (med) | All NSCLC | CT V (Xe-enhanced): >15 HU | VMAT IMRT |
Grade 2+ RP | Relative risk reduction compared with anatomic plan: 30%, P < .001 | Varian Eclipse’s biologic evaluations, using parameters from Seppenwoolde et al181 | Yes |
| Kanai96 | 40 | 77 (med) | Thoracic cancers | CT V (HU): Thresholds from 5th to 95th percentiles | SBRT | Grade 2+ RP | AUC: 0.57 | Best AUC found for fV30 at the 25th percentile threshold. Not statistically significant. | Yes |
| Kocak141 | 182 | N/A | 167 NSCLC 15 SCLC |
SPECT Q: Weighting | N/A | Grade 2+ RP | One-tailed Fisher’s exact P = .03 on original data set, P = .33 and .41 on other data sets AUC: 0.65 (bivariate, Duke) AUC: 0.72 (univariate/bivariate, Netherlands cancer institute (NKI)) |
Model predicts high risk if a patient has MLD ≥ 25 Gy and pre-RT DLCO less than (Overall perfusion-weighted response parameter (OpRP) + 38) AUC for Duke data set was OpRP and FEV1 AUC for NKI was OpRP or OpRP and DLCO or OpRP and FEV1 |
Variable depending on data set |
| Lan50 | 37 | 61 (mean) | 37 NSCLC | CT V (density change): Thresholded at 20%, 40%, 60% 80% of max; weighted | N/A | Grade 2+ radiation fibrosis | Lung consolidation: AUC: 0.66 (weighted) AUC: 0.65 (threshold) Volume loss: AUC: 0.71 (weighted) AUC: 0.75 (threshold) Statistically significant decrease in fV20, fV30, and fV40 for 60% thresholded parameters, for patients without volume loss Airway dilation: AUC: 0.8 (weighted) AUC: 0.85 (threshold) Statistically significant decrease in all thresholded functional parameters (60% threshold) and all but fV40 weighted parameters, for patients without airway dilation compared with those with |
Lung consolidation weighting model used fV20, and threshold used fV30 at a threshold of 20% Volume loss model used fV30 or fV40 for weighting and 40% threshold fV40, 60% threshold fV40, or 80% threshold fV40 for thresholded method Airway dilation model used fV20 for weighting and fV20 at 40% threshold for thresholded method |
Yes |
| Lee79 | 28 | 70.5 (med) | 14 NSCLC 5 SCLC 5 Locally recurrent 4 Lung mets |
SPECT Q: Threshold into 7 equidistant bins; 70% of max threshold used for dose-function parameters | VMAT proton IMRT SBRT 3DCRT |
Grade 2+ RP | rs = 0.94 AUC: 0.87, P =.011 pMLD cutoff 13.2 Gy EQD2: Sensitivity 100%; specificity 74% |
Best functional Spearman coefficient was for fV20 using perfusion Best AUC was for pF20 (perfusion-weighted) |
No |
| Li130 | 126 | 61 (med) | All lung cancer | CT Q: Model from Ren et al15 | IMRT | Grade 2+ RP | AUC: 0.862 (95% CI, 0.851-0.871) | Dual radiomics and perfusion image-based model | Yes |
| Li97 | 17 | 67 (med) | All NSCLC | CT V (Jacobian): Top 10%, 20%, 30%, 40%, and 50% of max (planning); weighting (dose-function parameter calculation) | IMRT | Grade 2+ RP | Previous NTCP model successfully predicted which population would have statistically significant improvements in functional parameters due to avoidance planning compared with anatomic plans. For this population only: 50% thresholded fV5 ↓, fV10 ↓, fV20 ↓, fMLD ↓, V5 ↓, V10 ↓. 40% thresholded fv5 ↓, fV10 ↓, fV20 ↓, fMLD ↓, V5 ↓. 40% thresholded fV5 ↓, fV20 ↓, fMLD ↓ This population also had: 50% thresholded PTV HI ↑, Cord Max ↑. 40% thresholded PTV HI ↑, cord max ↑. 30% thresholded HI max ↑, cord max ↑ |
NTCP model89 | Yes |
| Lind142 | 162 | 59 (mean) | 118 Lung 20 Breast 17 Lymphoma 8 Other |
SPECT Q: Weighting | N/A | Grade 2+ RP: Minimum 6 months follow-up | AUC: 0.62 (bivariate) AUC: 0.79 (bivariate with conditions) AUC: 0.83 (trivariate with conditions) |
Bivariate: DLCO and overall response parameter (ORP) Bivariate with conditions: DLCO > 40; Mean perfused lung dose (MPLD) and DLCO Trivariate with conditions: DLCO > 40; DLCO, FEV1, and MPLD or ORP |
Yes |
| Marks182 | 50 | N/A | 67 Lung 17 Breast 12 Lymphoma 4 Other |
SPECT Q: N/A | N/A | RT-related pulmonary symptoms | NTCPs based on functional parameters provided no additional predictive value | N/A | No |
| O’Reilly105 | 74 | N/A | All NSCLC | CT V (Jacobian): Top 6%, 45%, and 60% of max | Photon (unspecified) proton | Grade 2+ RP | AUC: 0.9 (photon) (95% CI, 0.74-0.98) AUC: 0.74 (proton) (95% CI, 0.53-0.89) |
Best AUC used fMLD in highly ventilated lung | Yes |
| Otsuka101 | 40 | 66 (med) | All thoracic cancer | CT V: Threshold at percentile regions from 0-100, in increments of 10 | SBRT other photon | Grade 2+ RP | AUC: 0.843 (95% CI, 0.732-0.954) For all functional parameters, grade 2+ RP had higher values than grade 1 RP |
Best AUC used fV5 to low-ventilated region (0-30th percentiles) | N/A |
| Owen106 | 88 | N/A | All NSCLC | SPECT V/Q: Percentile thresholds from 10-90, in increments of 10; analogous percentile thresholds of lowest function | 3DCRT VMAT |
Grade 2+ RP or clinical RF | Odds ratio: 0.05, P = .5 (Q) (95% CI, 0.01-0.91) Odds ratio: 0.06, P = .02 (V) (95% CI, 0.00-0.64) Odds ratio: 1.19, P = .006 (V and Q) |
Best for Q was normalized fV20 in ipsilateral lung. Best for V was normalized fV20 in both lungs. Best for V and Q combined was the low-function volume receiving ≥20 Gy |
No |
| Seppenwoolde181 | 382 | N/A | 274 NSCLC 66 Lymphoma 42 Breast |
SPECT Q: NA | N/A | Grade 2+ RP Grade 2+ RP |
Predictive; specific performance regarding quality of fit not specified. Log-likelihood of −119.4. | Sigmoid dose-effect relation fitted through SPECT Q | No |
| Sharifi78 | 30 | N/A | All NSCLC | CT V (Jacobian) CT V (volume): Threshold at 95% of max and weighting |
N/A | Grade 2+ dyspnea | Jacobian AUC: 0.79, P = .2 Volume AUC: 0.8, P = .01 |
Best AUC for Jacobian used fV1 to fV5 Best AUC for volume used fV1 to fV5 |
N/A |
| Thomas126 | 39 | 63 (med) | All NSCLC | SPECT Q: Radiomics, thresholding, and weighting separately each | IMRT VMAT PBT |
Grade 2+ RP | No functional parameters were found to be significantly predictive of the endpoint. | N/A | No |
| Vinogradskiy147 | 96 | N/A | All NSCLC | CT V (HU): Weighting | 3DCRT IMRT |
Grade 3+ RP | AUC: 0.62, P = .093 | Best AUC used fV20. None were significant. | Yes |
| Wang99 | 57 | N/A | All NSCLC | SPECT Q: Threshold 30% of max followed by weighting | IMRT 3DCRT |
Grade 2+ RP | RP rate stratification of 2.9%:43.5% AUC: 0.869, P = .0001 Sensitivity: 0.9091 Specificity: 0.7391 |
Best AUC used fV15 | Yes |
| Wang104 | 57 | N/A | All NSCLC | SPECT Q: Threshold 30% of max followed by weighting | 3DCRT IMRT |
Grade 2+ RILT | All functional parameters from fV5 to fV50 in 5 Gy increments were significantly higher in patients with RILT AUC:0.869, P = .001 (95%CI,0.764-0.973) |
Best AUC used fV15; best AUC better than volume-based fV40 | No |