State-Space Enhanced Grading of Prostate Cancer Pathological Images

Chaoyun Mai; Qianwen Wang; Runqiang Yuan; Zhipeng Mai; Chuanbo Qin; Junying Zeng; Weigang Yan; Yu Xiao

doi:10.3788/LOP242448

Laser & Optoelectronics Progress, Volume. 62, Issue 14, 1417004(2025)

State-Space Enhanced Grading of Prostate Cancer Pathological Images

Chaoyun Mai¹, Qianwen Wang¹, Runqiang Yuan^2、*, Zhipeng Mai³, Chuanbo Qin¹, Junying Zeng¹, Weigang Yan³, and Yu Xiao⁴

Author Affiliations

¹School of Electronics and Information Engineering, Wuyi University, Jiangmen 529020, Guangdong , China

²Department of Urology, Zhongshan City People's Hospital, Zhongshan 528403, Guangdong , China

³Department of Urology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing , 100730, China

⁴Department of Pathology, Peking Union Medical College Hospital, Peking Union Medical College, Chinese Academy of Medical Sciences, Beijing , 100730, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Overall network architecture of SKAN-MIL. (a) WSI preprocessing; (b) overall framework; (c) MP-Mamba module

Download full size

Fig. 2. KAN module

Download full size

Fig. 3. Division of training sets and test sets of different datasets. (a) PUMCH; (b) PANDA

Download full size

Fig. 4. ROC curve performances of SKAN-MIL model under different Gleason grades. (a) PUMCH; (b) PANDA

Download full size

Fig. 5. Accuracy radar charts for different methods on two datasets for each Gleason grade. (a) PUMCH; (b) PANDA

Download full size

Fig. 6. Confusion matrices for PUMCH and PANDA and normalized results. (a) Gleason grade confusion matrix of PUMCH; (b) Gleason grade confusion matrix of PANDA; (c) normalized results of PUMCH; (d) normalized results of PANDA

Download full size

Fig. 7. WSI heatmap visualization results on the PUMCH and PANDA

Download full size

Table 1. Number of WSI and the distribution of Gleason grade in the PUMCH and PANDA
View table
Table 1. Number of WSI and the distribution of Gleason grade in the PUMCH and PANDA
Gleason grade PUMCH PANDA
Benign 2303 2805
GG 1 302 2610
GG 2 675 1321
GG 3 173 1215
GG 4 53 1198
GG 5 269 1184
Total 3775 10333

Table 2. Comparison of classification performance of different methods on the PUMCH

View table

Table 2. Comparison of classification performance of different methods on the PUMCH

Method	AUC	Accuracy
Maxpool	88.28±0.49	79.63±0.31
Meanpool	83.87±1.17	75.54±0.43
ABMIL^［12］	86.51±0.48	74.06±0.26
CLAM_SB^［13］	85.54±0.86	80.32±0.91
CLAM_MB^［13］	78.23±2.36	72.89±2.06
DSMIL^［14］	87.83±0.19	76.76±0.36
TransMIL^［18］	88.43±1.39	76.61±1.41
Proposed	93.18±0.57	84.14±0.70

Table 3. Comparison of classification performance of different methods on the PANDA

View table

Table 3. Comparison of classification performance of different methods on the PANDA

Method	AUC	Accuracy
Maxpool	88.22±0.06	62.08±0.98
Meanpool	83.05±0.05	50.96±0.30
ABMIL^［12］	71.45±3.76	38.90±4.27
CLAM_SB^［13］	87.95±0.09	62.29±0.49
CLAM_MB^［13］	85.77±0.20	58.02±0.84
DSMIL^［14］	86.07±0.06	56.97±0.53
TransMIL^［18］	85.38±0.45	58.39±1.23
Proposed	89.50±0.33	63.92±1.19

Table 4. AUC of different methods for each Gleason grade on the PUMCH

View table

Table 4. AUC of different methods for each Gleason grade on the PUMCH

Method	Benign	GG 1	GG 2	GG 3	GG 4	GG 5
Maxpool	96.88±0.09	89.47±0.59	89.39±0.07	87.68±0.79	69.82±3.33	96.47±0.11
Meanpool	96.03±0.61	83.50±5.28	88.74±1.41	86.03±3.02	80.56±5.21	95.73±1.56
ABMIL^［12］	94.48±0.21	83.81±1.95	86.36±0.50	82.65±2.17	85.49±1.65	87.12±0.13
CLAM_SB^［13］	98.06±0.32	91.22±0.56	90.31±0.17	92.63±0.66	40.03±6.67	95.78±0.69
CLAM_MB^［13］	95.18±0.61	81.06±1.73	83.81±1.62	79.90±1.68	38.53±15.88	85.74±1.20
DSMIL^［14］	95.71±0.03	84.82±0.87	86.60±0.12	88.18±0.55	80.68±0.77	93.00±0.16
TransMIL^［18］	96.03±0.61	83.50±5.28	88.74±1.41	86.03±3.02	80.56±5.21	95.73±1.56
Proposed	98.11±0.68	89.28±1.41	91.18±2.27	93.18±2.27	87.29±3.58	97.19±1.63

Table 5. AUC of different methods for each Gleason grade on the PANDA

View table

Table 5. AUC of different methods for each Gleason grade on the PANDA

Method	Benign	GG 1	GG 2	GG 3	GG 4	GG 5
Maxpool	94.04±0.04	87.81±0.20	83.41±0.19	81.56±0.33	89.18±0.34	93.45±0.26
Meanpool	86.89±0.12	77.97±0.20	76.87±0.19	79.55±0.14	85.59±0.08	91.44±0.08
ABMIL^［12］	80.58±2.30	65.62±5.22	61.10±6.76	71.30±1.27	71.84±4.09	78.33±3.00
CLAM_SB^［13］	94.97±0.24	87.22±0.34	83.38±0.43	80.69±0.38	88.09±0.14	93.36±0.26
CLAM_MB^［13］	93.14±0.38	84.89±0.10	81.40±1.30	79.02±0.44	83.96±0.88	91.04±0.41
DSMIL^［14］	92.16±0.08	83.37±0.07	85.52±0.13	78.50±0.08	87.33±0.03	92.56±0.13
TransMIL^［18］	93.35±0.68	84.87±0.40	78.83±1.57	79.56±1.59	85.12±1.14	91.17±1.22
Proposed	95.81±0.26	89.59±0.29	85.81±1.13	83.42±1.05	89.10±0.74	93.84±0.55

Table 6. Ablation experiment of different modules on the PUMCH and PANDA
View table
Table 6. Ablation experiment of different modules on the PUMCH and PANDA
Model MP-Mamba KAN PUMCH PANDA
AUC Accuracy AUC Accuracy
1 × × 83.56±0.35 76.76±0.20 86.59±0.20 57.65±1.36
2 √ × 92.32±1.12 83.69±1.10 89.38±0.37 63.36±1.71
3 √ √ 93.18±0.57 84.14±0.70 89.50±0.33 63.92±1.19

Table 7. Experimental results of baseline model with different Mamba modules

View table

Table 7. Experimental results of baseline model with different Mamba modules

Module	PUMCH		PANDA
Module	AUC	Accuracy	AUC	Accuracy
Baseline	83.56±0.35	76.76±0.20	86.59±0.20	57.65±1.36
Mamba	90.49±0.89	82.30±0.10	88.68±0.20	62.49±1.54
Bi-Mamba	91.72±0.70	82.56±1.54	88.72±0.13	63.07±0.28
VMamba	91.06±0.80	82.78±1.10	88.90±0.25	63.12±1.25
MP-Mamba	92.32±1.12	83.69±1.10	89.38±0.37	63.36±1.71

Tools

Get Citation

Copy Citation Text

Chaoyun Mai, Qianwen Wang, Runqiang Yuan, Zhipeng Mai, Chuanbo Qin, Junying Zeng, Weigang Yan, Yu Xiao. State-Space Enhanced Grading of Prostate Cancer Pathological Images[J]. Laser & Optoelectronics Progress, 2025, 62(14): 1417004

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Medical Optics and Biotechnology

Received: Dec. 19, 2024

Accepted: Feb. 28, 2025

Published Online: Jul. 2, 2025

The Author Email: Runqiang Yuan (yuanrunqiang11@126.com)

DOI:10.3788/LOP242448

CSTR:32186.14.LOP242448

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Number of WSI and the distribution of Gleason grade in the PUMCH and PANDA

Table 1. Number of WSI and the distribution of Gleason grade in the PUMCH and PANDA

Table 2. Comparison of classification performance of different methods on the PUMCH

Table 2. Comparison of classification performance of different methods on the PUMCH

Table 3. Comparison of classification performance of different methods on the PANDA

Table 3. Comparison of classification performance of different methods on the PANDA

Table 4. AUC of different methods for each Gleason grade on the PUMCH

Table 4. AUC of different methods for each Gleason grade on the PUMCH

Table 5. AUC of different methods for each Gleason grade on the PANDA

Table 5. AUC of different methods for each Gleason grade on the PANDA

Table 6. Ablation experiment of different modules on the PUMCH and PANDA

Table 6. Ablation experiment of different modules on the PUMCH and PANDA

Table 7. Experimental results of baseline model with different Mamba modules

Table 7. Experimental results of baseline model with different Mamba modules