Comparative performance of fully-automated and semi-automated artificial intelligence methods for the detection of clinically significant prostate cancer on MRI: a systematic review

Sushentsev, Nikita; Moreira Da Silva, Nadia; Yeung, Michael; Barrett, Tristan; Sala, Evis; Roberts, Michael; Rundo, Leonardo

doi:10.1186/s13244-022-01199-3

Insights into Imaging

Table 5 Diagnostic performance of fully-automated and semi-automated AI methods for differentiating between csPCa and iPCa or benign disease

From: Comparative performance of fully-automated and semi-automated artificial intelligence methods for the detection of clinically significant prostate cancer on MRI: a systematic review

Study	Threshold	AUC [95% CI]	Accuracy	Sensitivity	Specificity	NPV	PPV
Studies using deep learning-based fully-automated AI algorithms
Wang [20]	NR	PZ: 0.89 [0.86–0.93] TZ: 0.97 [0.95–0.98]	PZ: 0.91 [0.86–0.95] TZ: 0.89 [0.87–0.91]	PZ: 0.60 [0.52–0.69] TZ: 1.0 [1.0–1.0]	PZ: 0.98 [0.95–1.0] TZ: 0.88 [0.82–0.93]	NR	NR
Fernandez-Quilez [21]	0.5	0.89	NR	0.85	0.94	NR	NR
Schelb [22]	Several for different PI-RADS cut-offs	NR	NR	PI-RADS ≥ 3: 0.96 PI-RADS ≥ 4: 0.92	PI-RADS ≥ 3: 0.31 PI-RADS ≥ 4: 0.47	PI-RADS ≥ 3: 0.84 PI-RADS ≥ 4: 0.83	PI-RADS ≥ 3: 0.53 PI-RADS ≥ 4: 0.67
Deniffel [23]	Risk of csPCa ≥ 0.2	0.85 [0.76–0.97]	NR	1.0 [1.0–1.0]	0.52 [0.32–0.68]	1.0 [1.0–1.0]	0.56 [0.48–0.66]
Seetharaman^a [24]	NR	0.80 (per lesion)	NR	0.70 (per lesion)	0.77 (per lesion)	NR	NR
Studies using traditional machine learning-based semi-automated AI algorithms
Bonekamp [25]	0.79	WP: 0.88 PZ: 0.84 TZ: 0.89 (per lesion)	NR	WP: 0.97 (per lesion)	WP: 0.58 (per lesion)	NR	NR
Min [26]	NR	0.82 [0.67–0.98]	NR	0.84	0.73	NR	NR
Castillo [28]	NR	0.75	NR	0.88	0.63	NR	NR
Bleker [29]	NR	0.87 [0.75–0.98]	NR	0.86	0.73	NR	NR
Woźnicki [31]	0.45	0.84 [0.6–1.0]	NR	0.91 [0.81–0.98]	0.57 [0.38–0.74]	NR	NR
Antonelli [34]	Reader SP (training)	PZ: 0.83 TZ: 0.75	NR	PZ: 90 TZ: 92	PZ: 65 TZ: 56	NR	NR
Hiremath [36]	Maximising accuracy (0.361)	0.81 [0.76–0.85]	0.78	0.83	0.59	NR	NR
Kwon^a [27]	NR	WP: 0.82	NR	NR	NR	NR	NR
Li^a [30]	− 0.42	0.98 [0.97–1.00]	0.90	0.95	0.87	0.97	0.82
Bevilacqua^a [32]	0.58	0.84 [0.63–0.90]	NR	0.9	0.75	NR	NR
Toivonen^a [33]	NR	0.88 [0.92–0.95]	NR	NR	NR	NR	NR
Yoo^a [35]	NR	0.84 [0.76–0.91]	NR	NR	NR	NR	NR

AUC, area under the receiver operating characteristic curve; NPV, negative predictive value; NR, not reported; PI-RADS, prostate imaging-reporting and data system; PPV, positive predictive value; PZ, peripheral zone; SP, specificity; TZ, transition zone; WP, whole prostate
^aThese papers had either high or unclear risk of bias on QUADAS-2 assessment (see Table 1; Fig. 2)

Back to article page