Attention-Guided Multi-Task Learning for Prostate Cancer Pelvic Lymph Node Metastasis Prediction

doi:10.16183/j.cnki.jsjtu.2023.491

Abstract

Abstract:

The prediction of pelvic lymph node metastasis (PLNM) based on quantitative preoperative magnetic resonance imaging features of prostate cancer primary tumor is an important reference for treatment planning. However, current prediction methods inadequately capture the heterogeneity within the primary tumor, resulting in a weak correlation between extracted quantitative image features and PLNM prediction. To address the aforementioned issues, an attention-guided multi-task learning network with tumor segmentation as an auxiliary task is proposed for PLNM prediction. First, within the tumor segmentation network, a multi-branch anisotropic large kernel attention module is introduced, where a larger receptive field is obtained through different branches and anisotropic large convolutional kernels, effectively capturing both local and global tumor information. Then, within the PLNM prediction network, a multi-scale feature interaction fusion attention module is introduced to hierarchically fuse and select features from multiple scales. The experimental results on a dataset of 320 cases demonstrate that the area under the precision-recall curve and the area under the receiver operating characteristic curve of the method proposed are (85.44±2.04)% and (91.86±2.18)%, which are superior to state-of-the-art methods and multi-task approaches.

Key words: prostate cancer pelvic lymph node metastasis (PLNM), multi-task learning, multi-branch anisotropic large kernel attention, multi-scale feature interaction fusion attention, multiparametric magnetic resonance imaging (mpMRI)

CLC Number:

ZHANG Zhiyuan, HU Jisu, ZHANG Yueyue, QIAN Xusheng, ZHOU Zhiyong, DAI Yakang. Attention-Guided Multi-Task Learning for Prostate Cancer Pelvic Lymph Node Metastasis Prediction[J]. Journal of Shanghai Jiao Tong University, 2025, 59(8): 1216-1224.

Figures/Tables 9

Fig.1

Fig.2

Fig.3

Fig.4

Tab.1

Tab.2

Fig.5

Tab.3

Tab.4

References 27

[1]	SIEGEL R L, MILLER K D, FUCHS H E, et al. Cancer statistics, 2022[J]. CA: A Cancer Journal for Clinicians, 2022, 72(1): 7-33.
[2]	WILCZAK W, WITTMER C, CLAUDITZ T, et al. Marked prognostic impact of minimal lymphatic tumor spread in prostate cancer[J]. European Urology, 2018, 74(3): 376-386. doi: S0302-2838(18)30423-8 pmid: 29908878
[3]	蒋璋栋, 吴开杰. 前列腺癌淋巴结转移分子机制的研究进展[J]. 现代泌尿外科杂志, 2022, 27(1): 79-83.
	JIANG Zhangdong, WU Kaijie. Progress of molecular mechanism leading to lymph node metastasis in prostate cancer[J]. Journal of Modern Urology, 2022, 27(1): 79-83.
[4]	YAKAR D, DEBATS O A, BOMERS J G R, et al. Predictive value of MRI in the localization, staging, volume estimation, assessment of aggressiveness, and guidance of radiotherapy and biopsies in prostate cancer[J]. Journal of Magnetic Resonance Imaging: JMRI, 2012, 35(1): 20-31.
[5]	扶昭, 陈燕清, 赵春丽, 等. 3.0 T MR扩散加权成像ADC值及ROC曲线分析对前列腺癌淋巴结转移的诊断价值[J]. 磁共振成像, 2021, 12(2): 49-51.
	FU Zhao, CHEN Yanqing, ZHAO Chunli, et al. Value of 3.0 T MR diffusion-weighted imaging and ROC curve in diagnosis of lymph node metastasis in patients with prostate cancer[J]. Chinese Journal of Magnetic Resonance Imaging, 2021, 12(2): 49-51.
[6]	BREMBILLA G, DELL’OGLIO P, STABILE A, et al. Preoperative multiparametric MRI of the prostate for the prediction of lymph node metastases in prostate cancer patients treated with extended pelvic lymph node dissection[J]. European Radiology, 2018, 28(5): 1969-1976. doi: 10.1007/s00330-017-5229-6 pmid: 29270644
[7]	PORPIGLIA F, MANFREDI M, MELE F, et al. Indication to pelvic lymph nodes dissection for prostate cancer: The role of multiparametric magnetic resonance imaging when the risk of lymph nodes invasion according to Briganti updated nomogram is 5[J]. Prostate Cancer and Prostatic Diseases, 2018, 21(1): 85-91.
[8]	FURUBAYASHI N, NEGISHI T, UOZUMI T, et al. Eliminating microscopic lymph node metastasis by performing pelvic lymph node dissection during radical prostatectomy for prostate cancer[J]. Molecular and Clinical Oncology, 2020, 12(2): 104-110. doi: 10.3892/mco.2019.1965 pmid: 31929879
[9]	LIU X, TIAN J Y, WU J Y, et al. Utility of diffusion weighted imaging-based radiomics nomogram to predict pelvic lymph nodes metastasis in prostate cancer[J]. BMC Medical Imaging, 2022, 22(1): 190. doi: 10.1186/s12880-022-00905-3 pmid: 36333664
[10]	ZHENG H X, MIAO Q, LIU Y K, et al. Multiparametric MRI-based radiomics model to predict pelvic lymph node invasion for patients with prostate cancer[J]. European Radiology, 2022, 32(8): 5688-5699. doi: 10.1007/s00330-022-08625-6 pmid: 35238971
[11]	HOU Y, BAO J, SONG Y, et al. Integration of clinicopathologic identification and deep transferrable image feature representation improves predictions of lymph node metastasis in prostate cancer[J]. EBioMedicine, 2021, 68: 103395.
[12]	ZHOU Y, CHEN H J, LI Y F, et al. Multi-task learning for segmentation and classification of tumors in 3D automated breast ultrasound images[J]. Medical Image Analysis, 2021, 70: 101918.
[13]	CHENG J H, LIU J, KUANG H L, et al. A fully automated multimodal MRI-based multi-task learning for glioma segmentation and IDH genotyping[J]. IEEE Transactions on Medical Imaging, 2022, 41(6): 1520-1532.
[14]	LAI H R, FU S R, ZHANG J, et al. Prior knowledge-aware fusion network for prediction of macrovascular invasion in hepatocellular carcinoma[J]. IEEE Transactions on Medical Imaging, 2022, 41(10): 2644-2657.
[15]	FUTREGA M, MILESI A, MARCINKIEWICZ M, et al. Optimized U-net for brain tumor segmentation[C]// International MICCAI Brainlesion Workshop. Cham: Springer, 2022: 15-29.
[16]	GUO M H, LU C Z, LIU Z N, et al. Visual attention network[J]. Computational Visual Media, 2023, 9(4): 733-752.
[17]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// Advances in Neural Information Processing Systems, New York, USA: Curran Associates Inc, 2017: 6000-6010.
[18]	CIPOLLA R, GAL Y, KENDALL A. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018: 7482-7491.
[19]	KLEIN S, STARING M, MURPHY K, et al. Elastix: A toolbox for intensity-based medical image registration[J]. IEEE Transactions on Medical Imaging, 2010, 29(1): 196-205. doi: 10.1109/TMI.2009.2035616 pmid: 19923044
[20]	ISENSEE F, JAEGER P F, KOHL S A A, et al. nnU-Net: A self-configuring method for deep learning-based biomedical image segmentation[J]. Nature Methods, 2021, 18(2): 203-211. doi: 10.1038/s41592-020-01008-z pmid: 33288961
[21]	SAITO T, REHMSMEIER M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets[J]. PLoS One, 2015, 10(3): e0118432.
[22]	LUQUE A, CARRASCO A, MARTÍN A, et al. The impact of class imbalance in classification performance metrics based on the binary confusion matrix[J]. Pattern Recognition, 2019, 91: 216-231.
[23]	LUO Y N, WANG Z S. An improved ResNet algorithm based on CBAM[C]// 2021 International Conference on Computer Network, Electronic and Automation. Xi’an, China: IEEE, 2021: 121-125.
[24]	HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017: 2261-2269.
[25]	TAN M X, LE Q V. EfficientNet: Rethinking model scaling for convolutional neural networks[C]// International Conference on Machine Learning, Long Beach, USA: PMLR, 2019: 6105-6114.
[26]	SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-ResNet and the impact of residual connections on learning[C]// Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence. San Francisco, USA: ACM, 2017: 4278-4284.
[27]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018: 7132-7141.

方法	(平均值±方差)/%
方法	AUPRC	AUROC	SEN	SPE	ACC	F₁
基线	78.91±6.07	90.11±3.21	67.02±10.42	91.27±2.66	84.37±3.99	70.73±7.94
基线+MBALKA	83.26±4.13	91.13±1.79	78.07±4.85	86.46±6.01	84.06±4.48	73.15±5.94
基线+MSFIFA	81.16±5.21	90.82±1.70	74.74±15.09	87.74±6.22	84.04±2.83	72.96±6.47
基线+MBALKA+MSFIFA	85.44±2.04	91.86±2.18	73.74±12.20	94.33±3.29	88.45±2.30	78.10±5.38

方法	(平均值±方差)/%
方法	AUPRC	AUROC	SEN	SPE	ACC	F₁
单任务	78.13±5.20	89.54±2.50	56.08±8.44	92.11±4.65	81.86±2.98	63.60±5.77
多任务 (等权重)	82.82±4.54	90.69±2.54	69.30±16.73	90.80±5.72	84.68±3.06	71.25±7.84
多任务 (不确定权重)	85.44±2.04	91.86±2.18	73.74±12.20	94.33±3.29	88.45±2.30	78.10±5.38

方法	(平均值±方差)/%
方法	AUPRC	AUROC	SEN	SPE	ACC	F₁
CBAMResNet50^[23]	78.83±5.01	89.33±3.68	58.25±12.11	90.79±5.12	81.55±4.44	63.88±9.64
DenseNet121^[24]	82.03±3.38	90.24±1.70	67.08±11.58	90.35±6.25	83.73±2.76	69.87±5.14
EfficientNet^[25]	80.10±6.68	88.25±3.84	64.74±5.55	87.36±11.00	80.93±8.67	66.82±10.87
InceptionV4^[26]	83.15±2.13	90.31±1.21	78.01±9.64	81.71±10.11	80.67±6.28	70.02±6.34
SeResNet50^[27]	83.30±4.80	90.30±3.18	73.63±2.41	92.14±4.50	86.88±3.58	76.34±5.13
本文方法	85.44±2.04	91.86±2.18	73.74±12.20	94.33±3.29	88.45±2.30	78.10±5.38

方法	(平均值±方差)/%
方法	AUPRC	AUROC	SEN	SPE	ACC	F₁
CMSVNet^[12]	82.68±4.53	89.81±2.42	81.29±8.49	80.32±6.49	80.62±2.44	70.49±1.49
MTTU-Net^[13]	81.80±6.53	91.65±2.41	84.62±11.37	85.07±11.00	84.96±5.79	76.58±5.30
PKAF-Net^[14]	78.68±5.16	88.52±2.03	69.12±11.76	91.29±5.08	85.00±2.10	72.09±4.67
多任务 (不确定权重)	85.44±2.04	91.86±2.18	73.74±12.20	94.33±3.29	88.45±2.30	78.10±5.38