Development of in-house fully residual deep convolutional neural network-based segmentation software for the male pelvic CT

Hirashima, Hideaki; Nakamura, Mitsuhiro; Baillehache, Pascal; Fujimoto, Yusuke; Nakagawa, Shota; Saruya, Yusuke; Kabasawa, Tatsumasa; Mizowaki, Takashi

doi:10.1186/s13014-021-01867-6

Table 1 Comparison of dataset, methodology (label and network), and the similarity scores (DSC and HD) reported by other studies and our study

From: Development of in-house fully residual deep convolutional neural network-based segmentation software for the male pelvic CT

Author	Number of datasets (patients)			Label	Network	Evaluation metrics	ROI				Commercial application
Author	Training	Validation	Test	Label	Network	Evaluation metrics	Prostate	Seminal vesicle	Rectum	Bladder	Commercial application
Macomber et al. [9]	94		99	Multiple	Deep decision forests	DSC [median (IQR)]	0.75 (0.67–0.82)	0.49 (0.31–0.79)	0.71 (0.63–0.87)	0.94 (0.92–0.98)
Macomber et al. [9]	94		99	Multiple	Deep decision forests	HD [mm]	–	–	–	–
Balagopal et al. [10]	136 (including tests)			Multiple	ResNeXt (3D-Unet)	DSC (mean ± SD)	0.90 ± 0.20	–	0.84 ± 0.37	0.95 ± 0.15
Balagopal et al. [10]	136 (including tests)			Multiple	ResNeXt (3D-Unet)	HD [mm]	–	–	–	–
Liu et al. [11]	771	193	140	Single	Deep neural network	DSC (mean ± SD, range)	0.85 ± 0.06 (0.65–0.93)	–	–	–
Liu et al. [11]	771	193	140	Single	Deep neural network	HD [mm] (mean ± SD)	7.0 ± 3.5	–	–	–
Zhang et al. [12]	90	10	20	Multiple	ARPM-Net	DSC (mean ± SD)	0.88 ± 0.11	–	0.86 ± 0.12	0.97 ± 0.07
Zhang et al. [12]	90	10	20	Multiple	ARPM-Net	Average HD [mm] (mean ± SD)	1.58 ± 1.77	–	3.14 ± 2.39	1.91 ± 1.29
Wang et al. [13]	268		45	Multiple	U-net	DSC (mean ± SD)	0.89 ± 0.03	–	0.89 ± 0.04	0.94 ± 0.03
Wang et al. [13]	268		45	Multiple	U-net	HD [mm]	–	–	–	–
Kijunen et al. [14]	876		30	Multiple	3D U-net	DSC (mean)	0.82	0.72	0.84	0.93
Kijunen et al. [14]	876		30	Multiple	3D U-net	HD [mm] (mean)	6.1	7.1	11.4	3.3
Czeizler et al. [19]	87		5	Multiple	BibNet	DSC (mean ± SD)	–	–	0.75 ± 0.11	0.90 ± 0.06
Czeizler et al. [19]	87		5	Multiple	BibNet	HD [mm]	–	–	–	–
Schreier et al. [20]	300		50	Multiple	BibNet	DSC (mean)	0.84	0.70	0.87	0.93
Schreier et al. [20]	300		50	Multiple	BibNet	HD [mm]	–	–	–	–
Wong et al. [24]	328		50	Multiple	U-net	DSC (minimum)	0.79	0.64	0.78	0.97	Limbus Contour
Wong et al. [24]	328		50	Multiple	U-net	95%HD [mm]	6.72	5.95	12.09	3.24
Our study	270	90	90	Multiple	FusionNet	DSC [median (IQR)]	0.87 (0.85–0.89)	0.77 (0.69–0.82)	0.91 (0.87–0.92)	0.96 (0.94–0.97)
			90	Multiple	FusionNet	HD [mm]	–	–	–	–
			20	Multiple	FusionNet	DSC [median (IQR)]	0.82 (0.79–0.84)	0.71 (0.67–0.77)	0.89 (0.86–0.91)	0.95 (0.94–0.96)
			20	Multiple	FusionNet	95%HD [mm] [median (IQR)]	3.23 (2.99–3.42)	3.82 (3.49–4.29)	2.65 (2.39–2.92)	4.18 (3.52–4.77)

DSC dice similarity coefficient, HD Hausdorff distance, 95%HD 95th-percentile Hausdorff distance, IQR interquartile range, ROI region of interest

Back to article page

ISSN: 1748-717X

Contact us

General enquiries: journalsubmissions@springernature.com

Radiation Oncology

Contact us