Abstract
Pancreatic cancer is often diagnosed at an advanced stage, making accurate imaging important for lesion assessment and treatment monitoring. Diffusion-weighted MRI (DWI) is a promising tool for probing tissue microstructure, but quantitative pancreatic DWI remains limited by respiratory and physiological motion, low SNR, and unstable high-b-value measurements. We propose a self-supervised physics-guided implicit neural representation (INR) for subject-specific pancreatic DWI reconstruction and biomarker estimation. Rather than denoising image intensities directly, the INR predicts continuous diffusion parameter fields and reconstructs DWI volumes through a signal-decay model, enforcing b-value-consistent behavior without clean references. We evaluated abdominal DWI from six subjects, including three confirmed pancreatic cancer cases, acquired at , , , and . INR reconstruction was compared with scanner reconstruction and MP-PCA-processed DWI using visual assessment and ROI-based analysis of apparent diffusion coefficient (ADC), synthetic ADC (sADC), and distributed diffusion coefficient (DDC) biomarkers. INR produced smoother high-b-value images with more coherent anatomical boundaries. In a single-reader qualitative assessment, INR received the highest mean scores among the three reconstruction methods for overall image quality and lesion conspicuity (overall composite 2.77, versus 2.43 for MP-PCA and 1.90 for the scanner). Cancer tissue generally showed lower diffusion biomarker values than normal tissue, with the clearest separation observed for sADC and DDC after INR reconstruction. These preliminary results suggest that physics-guided INR may improve pancreatic DWI interpretability and quantitative biomarker estimation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CaPTion_010.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/profile?id=~Nitzan_Avidan-Pearl1
BibTex
@InProceedings{AviNit_PhysicsGuided_MICCAISAT2026,
author = { Avidan-Pearl, Nitzan AND Link-Sourani, Daphna AND Freiman, Moti AND Ogawa, Hiroshi AND Yamao, Kentaro AND Iida, Tadashi AND Takami, Hideki AND Iima, Mami},
title = { { Physics-Guided Implicit Neural Representations for Enhanced Quantitative DWI Biomarkers in Pancreatic Cancer } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}