MRI order improves prediction of nerve invasion risk in liver cancer

Order-Aware 2.5D Multiple Instance Learning for Preoperative MRI-Based Perineural Invasion Risk Assessment in Intrahepatic Cholangiocarcinoma

Computer Vision and Pattern Recognition

Summary

Perineural invasion is a harmful sign found in a type of liver cancer but usually detected only after surgery. The authors developed a method to predict this risk before surgery using MRI scans by analyzing sequences of images in a specific order. Their approach uses a special machine learning technique that looks at sequences of overlapping image slices and combines information to guess the risk. This method did better than other approaches that ignored the image order. It shows that considering the order of MRI images helps identify nerve invasion risks without needing detailed labels for every image slice.

What this means in practice

  • For radiology teams: Predict the risk of nerve invasion in liver cancer patients before surgery using ordered MRI image analysis to assist treatment planning.
  • For medical imaging software developers: Incorporate sequence-aware learning models to improve patient-level predictions from MRI scans without detailed image annotations.

Authors

Hyunsu Go, Youngung Han, Kyeonghun Kim, Jinyong Jun, Junbeom Lee, Dohyun Kweon, Yului Jeong, Suah Park, Sungha Park, Anna Jung, Woo Kyoung Jeong, Ken Ying-Kai Liao, Hyuk-Jae Lee, Nam-Joon Kim

Abstract

Perineural invasion (PNI) is an adverse histopathologic marker in intrahepatic cholangiocarcinoma (ICC), but it is usually confirmed only after resection. Preoperative T2-weighted MRI may provide noninvasive imaging cues predictive of PNI, although labels are available only at the patient level without slice- or voxel-level annotations. We propose Order-Aware Slab Multiple Instance Learning (OAS-MIL), a weakly supervised framework for patient-level PNI prediction. Each tumor-centered MRI crop is represented as an ordered sequence of overlapping 2.5D slabs formed from contiguous axial slices. A shared encoder extracts slab-level features, which are aggregated by a permutation-invariant set-attention branch and a bidirectional sequence-attention branch. Using five-fold label-stratified cross-validation at the patient level, OAS-MIL achieved a mean AUROC of 0.770, outperforming the evaluated volumetric and MIL baselines. These results suggest that axial order provides a useful inductive bias for weakly supervised PNI prediction from MRI.
View PDFOpen arXiv