Consistency in contouring of organs at risk by artificial intelligence vs oncologists in head and neck cancer patients

Figures & data

Table 1. DSC and MSD for 12 OARs contoured by AI oncologists, respectively.

	DSC AI		DSC oncologists		MSD AI		MSD oncologists
	[index]	IQR	[index]	IQR	[mm]	IQR	[mm]	IQR
Extended oral cavity	0.95	0.94 − 0.95	0.84	0.81 − 0.89	0.8	0.7 − 1.0	2.2	1.7 − 2.9
PCM up	0.79	0.76 − 0.81	0.40	0.31 − 0.51	0.8	0.7 − 0.9	2.4	2.0 − 3.0
PCM mid	0.76	0.72 − 0.79	0.51	0.41 − 0.59	0.8	0.7 − 0.9	2.0	1.5 − 2.6
PCM low	0.75	0.70 − 0.79	0.50	0.39 − 0.60	0.9	0.7 − 1.1	2.3	1.7 − 3.5
Glottic larynx	0.79	0.76 − 0.82	0.45	0.35 − 0.57	0.8	0.6 − 0.9	2.2	1.7 − 3.0
Suppraglottic larynx	0.88	0.86 − 0.89	0.58	0.48 − 0.64	0.9	0.8 − 1.1	1.9	1.6 − 2.6
Parotid left	0.91	0.89 − 0.92	0.82	0.77 − 0.85	0.9	0.8 − 1.1	1.7	1.6 − 2.3
Parotid right	0.93	0.92 − 0.93	0.82	0.78 − 0.84	0.8	0.7 − 0.9	1.7	1.5 − 2.1
Submandibular left	0.86	0.83 − 0.88	0.78	0.74 − 0.82	1.0	0.9 − 1.2	1.4	1.1 − 1.8
Submandibular right	0.89	0.86 − 0.90	0.78	0.74 − 0.81	0.8	0.8 − 1.0	1.5	1.2 − 1.9
Thyroid	0.83	0.78 − 0.86	0.72	0.67 − 0.78	1.2	0.9 − 1.4	1.7	1.4 − 2.2
Oesophagus	0.81	0.78 − 0.85	0.63	0.55 − 0.71	1.1	0.9 − 1.6	3.3	2.1 − 6.2
Oesophagus*	0.84	0.81 − 0.86	0.72	0.67 − 0.77	0.8	0.7 − 1.0	1.9	1.3 − 2.6

Oesophagus*: comparison of oesophagus contour overlap after correction so the contours have the same caudal length. PCM up: upper pharyngeal constrictor muscle, PCM Mid: middle pharyngeal constrictor muscle, PCM low: lower pharyngeal constrictor muscle. Interquartile range (IQR).

Figure 1. Box plot with individual samples overlaid showing the DSC for the 12 OARs. Green boxes and samples show the DSC for the AI contours, and blue boxes and samples are results comparing oncologist contours. The raw data points are shown to visualise the distribution.

Figure 2. Box plot with individual samples overlaid showing the MSD for the 12 OARs. Green boxes and samples show the MSD for the AI contours, and blue boxes and samples are results comparing oncologist contours. The raw data points are shown to visualise the distribution. For visualisation, the plot has been scaled, omitting two outliers from oesophagus and one from glottic larynx for oncologist contours.

Figure 3. Scatter plot of the local ΔNTCP ($\mathit{NTC}{P}_{\mathit{local}\mathrm{}\mathit{photon}\mathrm{}\mathit{plan}}-\mathit{NTC}{P}_{\mathit{local}\mathrm{}\mathit{proton}\mathrm{}\mathit{plan}}$) and clinical ΔNTCP ($\mathit{NTC}{P}_{\mathit{local}\mathrm{}\mathit{photon}\mathrm{}\mathit{plan}}-\mathit{NTC}{P}_{\mathit{clinical}\mathrm{}\mathit{proton}\mathrm{}\mathit{plan}}$) based on AI (green data points) and oncologist (blue data points) contours, respectively, for xerostomia and dysphagia.

Figure 4. Bland-Altman plot showing the mean and difference between the local and clinical $\Delta$NTCP for xerostomia and dysphagia. The green data points represent the $\Delta$NTCP based on AI contours, and the blue data points represent the $\Delta$NTCP calculated based on oncologist contours.

Data availability statement

The data used in this study are part of a clinical trial and are not available.