Abstract

Nucleus instance segmentation is essential for cancer diagnosis, yet existing methods primarily follow a segment-all-then-classify paradigm lacking mechanisms for selective analysis. We present Conditional Instruction-guided Pathology Segmentation Network (CIPS-Net), a text-conditional framework allowing pathologists to segment specific nucleus types via natural language instructions. Our architecture combines a DINOv2 ViT-B/14 vision encoder with Bio_ClinicalBERT text encoding, fusing features through cross-attention and language-modulated skip connections. A four-head design jointly predicts nucleus presence, distance transforms, instance embeddings, and type classification. To encourage conditional behaviour without additional annotations, we introduce permutation-based training that enumerates all 2N −1 non-empty class subsets per image. On PanNuke, CIPS-Net achieves an mPQ of 0.4846 and a bPQ of 0.6217 (3-fold cross-validation), remaining competitive with unconditional SOTA methods while supporting text-conditional selective segmentation at higher inference speeds (32.3 FPS) and lower computational cost (52.6 G MACs) than the top-performing unconditional baseline. Notably, CIPS-Net yields competitive Dead-class PQ (0.211) and Inflammatory PQ (0.471) among evaluated methods, suggesting that text conditioning may benefit the analysis of rare, diagnostically critical cell types.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/COMPAYL_019.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: https://openreview.net/forum?id=WSozvtw6Yv&referrer=%5BProgram%20Chair%20Console%5D(%2Fgroup%3Fid%3DMICCAI.org%2F2026%2FWorkshop%2FCOMPAYL%2FProgram_Chairs%23submission-status)

BibTex

@InProceedings{SulNik_CIPSNet_MICCAISAT2026,
        author = { Sulake, Nikhileswara Rao AND Machara, Sai Manikanta Eswar AND Retta, Sivaji AND Ganapathi, Iyyakutti Iyappan AND Owais, Muhammad AND Hussain, Irfan},
        title = { { CIPS-Net: Text-Instructed Conditional Nucleus Instance Segmentation Network in Histopathology } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17251},
        month = {pending},
        page = {pending}
}


back to top