Prediction of plan adaptation in head and neck cancer proton therapy using clinical, radiographic, and dosimetric features

Figures & data

Table 1. Patient data.

	No Re-Plan (%)	Re-Plan (%)	Total	p-value
Stage				.419
^aLocally Advanced	66.4	33.6	131
Metastatic	100	0	1
Recurrent	100	0	4
^bEarly	70.5	29.5	61
Other	85.7	14.3	14
Tumor Site				.026
Nasal Cavity and Sinuses	61.5	38.5	26
Hypopharynx/Larynx	60.0	40.0	8
Lip and Oral Cavity	76.0	24.0	25
Unknown Primary	93.8	6.2	16
Salivary Glands	100	0	17
Nasopharynx	53.3	46.7	15
Oropharynx	56.6	43.4	53
Sarcoma	76.2	23.8	21
^cOther	77.7	22.3	18
T Values				.185
0	63.2	36.8	19
1	87.5	12.5	24
2	72.7	27.3	44
3	65.0	35.0	40
4	56.7	43.3	30
4a	71.4	28.6	35
4b	61.5	38.5	13
N Values				.157
0	70.2	29.8	94
1	78.7	21.3	47
2	59.0	41.0	39
3	58.3	41.7	24
M Values				.668
0	66.5	33.5	200
1	60.0	40.0	5
P16 Values				.574
N/A	72.0	28.0	125
Negative	71.0	29.0	31
Positive	64.2	35.8	53
Surgery Status				.001
Had Surgery	82.2	17.8	101
No Surgery	58.2	41.8	110
Surgery Margins				.002
N/A	59.7	40.3	124
Negative	77.5	22.5	40
Close	87.0	13.0	23
Positive	91.3	8.7	23
Treatment Status				.606
Palliative	80.0	20.0	5
Definitive	71.2	28.8	205
Chemo Status				.001
Had Chemotherapy	56.3	43.7	119
No Chemotherapy	87.0	13.0	92
Gender				.613
Male	71.7	28.3	127
Female	66.7	33.3	84
Age (Years)				.207
Minimum	2.0	2.0
Maximum	96.0	94.0
Mean	61.6	58.2
Median	65.0	62.0
Treatment Fractions				.000
Minimum	5.0	4.0
Maximum	37.0	37.0
Mean	30.1	32.6
Median	30.0	35.0

Notes: Varying counts in total patients between variables is due to missing data for certain patients.

^aLocally advanced stages include III-IVc.

^bEarly include stages I-II.

^cOther tumor sites include the central nervous system, cutaneous, hematologic malignancies, and thoracic.

Figure 1. (a) Isodose for a typical bilateral HN patient. (b) DVH for all 3 CTVs of the HN patient. (c) Isodose on a robustness evaluation scenario. (d.1) DVH of the CTVs for the nominal plan (dashed) and the robustness scenario (solid). (d.2) Tail of the CTV DVH. Mean max dose is calculated from the tail of the DVH. (e) Hotspot and dose heterogeneity from the 330° right posterior oblique (RPO) beam. (f) DVH for all 3 CTVs from the RPO beam. Notes: There is a > 5% max dose change in (d). this is shown in the isodose with the pink hotspots in (c). Figure (e)’s white hotspot represents a 110% beam dose heterogeneity.

Figure 2. Multilayer perceptron neural network for re-plan prediction. BHI: beam heterogeneity index, postop: whether the patient had surgery, dmax variation: max dose change across robustness scenarios, chemo: whether the patient is undergoing chemotherapy, h: hidden unit.

Figure 3. (a) Distributions of mean beam heterogeneity index (BHI) and success rate. (b) Distributions of mean V100 and max dose change. (c) Normalized importance for all features used in the prediction model. (d) ROC curve for the prediction model. The area under the curve (AUC) is .855.

Table 2. Results for numerical features.

	Count	Mean	Interquartile Range	p-value
Mean BHI				.005
No Re-Plan	103	0.54	0.05
Re-Plan	57	0.56	0.05
Success Rate (%)				.785
No Re-Plan	103	57.0	57.3
Re-Plan	57	52.0	57.1
Mean V100 (%)				.272
No Re-Plan	103	94.51	3.10
Re-Plan	57	94.15	3.20
Mean V100 Change (%)				.638
No Re-Plan	103	3.85	2.90
Re-Plan	57	4.00	2.50
Mean Max Dose Change (%)				.008
No Re-Plan	103	2.95	2.00
Re-Plan	57	3.15	1.00

Notes: Mean BHI and robustness features come from the same export script. Different total patient counts than Table 1 are because of varying export script passing rates and outlier removal.

Data sharing

Due to ethical/legal/commercial reasons, supporting data is not available.

Data availability statement

Research data are not available at this time.