notesum.ai

Published at November 18

Cross-Patient Pseudo Bags Generation and Curriculum Contrastive Learning for Imbalanced Multiclassification of Whole Slide Image

cs.CV

cs.AI

Released Date: November 18, 2024

Authors: Yonghuang Wu¹, Xuan Xie¹, Xinyuan Niu¹, Chengqian Zhao¹, Jinhua Yu¹

Aff.: ¹Fudan University

Arxiv: http://arxiv.org/abs/2411.11262v1

Refer to caption

(a) TCGA Kidney
Methods	F1	AUC	ACC	MCC	SEN	SPE	PPV	NPV
DSMIL	66.24	95.47	80.00	58.09	66.45	87.79	77.60	89.48
TransMIL	73.78	95.27	84.04	68.16	72.55	90.77	86.64	91.77
CLAM	64.75	94.84	79.55	55.81	65.68	87.65	70.57	89.33
DTFD	57.76	94.45	82.13	50.62	62.07	89.26	54.22	91.81
Ours	84.21	97.47	88.88	79.28	82.40	93.84	89.46	94.11
(b) Liver Cancer
Methods	F1	AUC	ACC	MCC	SEN	SPE	PPV	NPV
DSMIL	87.42	99.13	92.64	84.43	92.70	96.99	85.87	94.59
TransMIL	70.24	94.32	84.56	62.41	66.88	87.19	85.03	91.32
CLAM	84.24	98.92	92.12	80.58	85.10	94.77	87.48	95.44
DTFD	32.94	92.17	56.27	18.97	41.80	72.64	37.67	71.23
Ours	88.95	98.78	93.78	85.54	92.47	97.07	86.94	95.15
(c) Camelyon 17
Methods	F1	AUC	ACC	MCC	SEN	SPE	PPV	NPV
DSMIL	74.80	80.14	80.00	50.87	76.86	76.86	74.14	74.14
TransMIL	77.89	86.54	82.20	57.27	80.88	80.88	76.59	76.59
CLAM	72.73	87.02	76.00	51.19	79.36	79.36	72.31	72.31
DTFD	80.04	80.92	85.80	61.01	78.97	78.97	82.29	82.29
Ours	80.86	84.63	85.20	62.66	82.00	82.00	80.82	80.82