[KongchangAI]
ModelCLEAR-WSI

CLEAR-WSI

A diagnosis-driven retrieval framework for pathology whole-slide images. It represents an entire slide using a fixed-length embedding and combines diagnostic label refinement with a scalable memory bank to achieve efficient retrieval.

Timeline (last 90 days)

Nov 6

CLEAR-WSI is a scalable, diagnosis-driven whole-slide image retrieval framework that uses fixed-length embeddings to represent an entire WSI.

Unverified50%
Nov 6

In the retrieval stage, CLEAR-WSI uses predicted diagnostic labels or ground-truth labels, depending on the retrieval setting, to refine the candidate results.

Unverified50%
Nov 6

CLEAR-WSI uses a scalable memory bank, and newly processed cases are added in an append-only manner without retraining or rebuilding the index.

Unverified50%
Nov 6

The study evaluated CLEAR-WSI on a comprehensive benchmark comprising two public datasets (CAMELYON16, BRACS) and one private clinical dataset (X-HER2), covering four retrieval tasks.

Unverified50%
Nov 6

Compared with the strongest baseline, CLEAR-WSI improves Precision@5 across the board: CAMELYON16 by 3.10, BRACS-1 by 2.75, BRACS-2 by 1.95, and X-HER2 by 3.75.

Unverified50%
Nov 6

The improvement on the X-HER2 task is the largest, possibly suggesting that diagnosis-driven refinement has more of an advantage in specific subtype classification scenarios.

Unverified50%
Nov 6

The Precision@5 improvements are at the single-digit level, still far from the reliability required for clinical deployment.

Unverified50%
Nov 6

Although the latency study is mentioned, the specific figures are not given in the abstract, and the actual cost of scaled deployment still needs further evaluation.

Unverified50%

All Facts (8)

Source Articles