DiffInfinite: large mask-image synthesis via parallel random patch diffusion in histopathology

Aversa, Marco; Nobis, Gabriel; Hägele, Miriam; Standvoss, Kai; Chirica, Mihaela ; Murray-Smith, Roderick; Alaa, Ahme; Ruff, Lukas; Ivanova, Daniel; Samek, Wojciech; Klauschen, Frederick ; Sanguinetti, Bruno und Oala, Luis (2023): DiffInfinite: large mask-image synthesis via parallel random patch diffusion in histopathology. 37th Conference on Neural Information Processing Systems (NeurIPS), New Orleans, USA, 10. - 16. Dezember 2023. Oh, Alice (Hrsg.): In: Advances in Neural Information Processing Systems 36 (NeurIPS 2023), Advances in Neural Information Processing Systems; 36 3413 New York: Curran Associates. S. 78126-78141

Abstract

We present DiffInfinite, a hierarchical diffusion model that generates arbitrarily large histological images while preserving long-range correlation structural information. Our approach first generates synthetic segmentation masks, subsequently used as conditions for the high-fidelity generative diffusion process. The proposed sampling method can be scaled up to any desired image size while only requiring small patches for fast training. Moreover, it can be parallelized more efficiently than previous large-content generation methods while avoiding tiling artifacts. The training leverages classifier-free guidance to augment a small, sparsely annotated dataset with unlabelled data. Our method alleviates unique challenges in histopathological imaging practice: large-scale information, costly manual annotation, and protective data handling. The biological plausibility of DiffInfinite data is evaluated in a survey by ten experienced pathologists as well as a downstream classification and segmentation task. Samples from the model score strongly on anti-copying metrics which is relevant for the protection of patient data.

• Dokumententyp: Konferenzbeitrag (Paper)
• Fakultät: Medizin > Pathologisches Institut
• Themengebiete: 600 Technik, Medizin, angewandte Wissenschaften > 610 Medizin und Gesundheit
• ISBN: 978-1-7138-9992-1
• Ort: New York
• Sprache: Englisch