Automatic segmentation of prostate zonal anatomy on MRI: a systematic review of the literature

Wu, Carine; Montagne, Sarah; Hamzaoui, Dimitri; Ayache, Nicholas; Delingette, Hervé; Renard-Penna, Raphaële

doi:10.1186/s13244-022-01340-2

Insights into Imaging

Table 5 Overview of segmentation methods with performance based on DSC. Number of articles reporting stratification by gland height, and reporting pre- or post-processing steps

From: Automatic segmentation of prostate zonal anatomy on MRI: a systematic review of the literature

First author, year of publication	Type	DSC results^†				Stratification by gland height	Pre-processing details	Post-processing details
First author, year of publication	Type	WG	TZ	PZ	CG	Stratification by gland height	Pre-processing details	Post-processing details
Cuocolo et al. [43]	CNN	0.9063*	–	0.7142*	0.8692*	×	✓	×
Bardis et al. [42]	CNN	0.94	0.91	0.774	–	×	✓	×
Lai et al. [41]	CNN	–	0.93	0.7004	–	×	✓	×
Nai et al. [18]	CNN	0.89*	–	0.712*	0.856*	✓	✓	×
Sanford et al. [40]	CNN	0.915	0.89	–	–	×	✓	×
Aldoj et al. [39]	CNN	0.921*	–	0.781*	0.895*	✓	✓	×
Zavala-Romero et al. [6]	CNN	0.825^a 0.892^b	–	0.788 ^a 0.811 ^b	–	×	✓	✓
Lee et al. [38]	CNN	0.8712	0.7648	–	–	×	✓	×
Liu et al. [37]	CNN	–	0.89^c 0.87^d	0.80^c 0.79^d	–	✓	✓	×
Qin et al. [36]	CNN	–	–	0.806	0.901	×	✓	✓
Motamed et al. [35]	CNN	0.89^e 0.85^f	0.86^e 0.84^f	–	–	×	×	✓
Zabihollahy et al. [13]	CNN	0.9533^g 0.9209^h	–	0.8678^g 0.861^h	0.9375^g 0.8989^h	✓	✓	✓
Padgett et al. [8]	Atlas	0.83*	0.75*	0.59*	–	✓	×	×
Rundo et al. [15]¹	CNN	–	–	0.919ⁱ 0.831^j 0.801^k	0.871ⁱ 0.886^j 0.937^k	×	✓	✓
Meyer et al. [34]	CNN	–	0.876	0.798	–	×	✓	✓
Liu et al. [16]	CNN	–	0.86^c 0.79^d	0.74^c 0.74^d	–	✓	✓	×
Rundo et al. [33]²	CNN	–	–	0.91* (with pre-training)	0.85* (with pre-training)	×	✓	✓
Hambarde et al. [32]	CNN	–	–	0.8733	–	×	✓	×
Jensen et al. [31]	CNN	–	–	0.692	0.794	✓	✓	✓
Khan et al. [17]	CNN	–	–	0.703*	0.88*	×	×	×
Cheng et al. [30]	CNN	0.9235*	–	–	0.9006*	✓	✓	✓
Zhu et al. [12]	CNN	0.927	–	0.793	–	✓	✓	×
Mooij et al. [29]	CNN	–	0.85*	0.6*	–	×	✓	×
Can et al. [28]	CNN	–	–	0.722*	0.89*	×	×	×
Clark et al. [14]	CNN	0.886^c 0.862^d	0.847^c	–	–	×	✓	×
Chilali et al. [9]	C means + Atlas	0.9478	0.7023	0.62	–	×	✓	×
Makni et al. [10]	C means	–	0.88	0.78	–	×	✓	×
Chi et al. [27]	Gaussian model	0.8	–	0.53	0.83	×	×	×
Toth et al. [26]	Active appearance model	0.81	–	0.68^l 0.60^m	0.79^l 0.72^m	✓	✓	×
Litjens et al. [7]	Atlas	–	–	0.75	0.8	×	×	×
Moschidis and Graham [25]	Random Forrest + Graph Cuts	–	–	–	–	×	✓	×
Yin et al. [24]	Graph Cuts	–	–	–	0.81	×	×	×
Makni et al. [23]	C means	–	–	0.76^l	0.87^l	✓^¶	×	×

CNN convolutional neural network
^†Dice similarity coefficient (DSC) for whole gland (WG), transition zone (TZ), peripheral zone (PZ) or central gland (CG) (means)
*Best results if several models were tested
^¶no Dice Similary Coefficien (DSC) provided
^a,bTrained on combined datasets and, respectively, tested on GE^a or Siemens^b dataset
^c,dRespectively for testing on internal^c or external^d data
^e,fRespectively for source^e or target^f with 115 patients for training (best results)
^g,hRespectively for T2-weighted^g and apparent diffusion coefficient (ADC) map^h
^I,j,kTrained on combined datasets and, respectively, tested on dataset #1ⁱ, #2^j or #3^k
^l,mUsing pre-segmented whole gland (WG)^l, or with whole process^m
¹Rundo et al., USE-Net: incorporating Squeeze-and-Excitation blocks into U-Net for prostate zonal segmentation of multi-institutional MRI datasets[30]
²Rundo et al., CNN-based Prostate Zonal Segmentation on T2-weighted MR Images: A Cross-dataset Study[27]

Back to article page