A Self-Supervised Image Registration Approach for Measuring Local Response Patterns in Metastatic Ovarian Cancer

TL;DR

High-grade serous ovarian carcinoma exhibits significant spatial and temporal heterogeneity, complicating monitoring of tumour burden during neoadjuvant chemotherapy. The authors propose a self-supervised deformable image registration framework based on GradICON/uniGradICON encoders to co-register pre- and post-NACT CE-CT scans and compute Jacobian maps that quantify voxel-wise and tissue-subregion changes across multiple disease sites. They show improved registration accuracy over NiftyReg and VoxelMorph and reveal heterogeneous responses between omental and pelvic disease and among hypo-/hyper-dense regions, suggesting potential as a sensitive marker for long-term pathological response and survival. This approach enables lesion-wise, three-dimensional assessment of treatment effects beyond RECIST, with implications for personalised treatment planning in advanced ovarian cancer.

Abstract

High-grade serous ovarian carcinoma (HGSOC) is characterised by significant spatial and temporal heterogeneity, typically manifesting at an advanced metastatic stage. A major challenge in treating advanced HGSOC is effectively monitoring localised change in tumour burden across multiple sites during neoadjuvant chemotherapy (NACT) and predicting long-term pathological response and overall patient survival. In this work, we propose a self-supervised deformable image registration algorithm that utilises a general-purpose image encoder for image feature extraction to co-register contrast-enhanced computerised tomography scan images acquired before and after neoadjuvant chemotherapy. This approach addresses challenges posed by highly complex tumour deformations and longitudinal lesion matching during treatment. Localised tumour changes are calculated using the Jacobian determinant maps of the registration deformation at multiple disease sites and their macroscopic areas, including hypo-dense (i.e., cystic/necrotic), hyper-dense (i.e., calcified), and intermediate density (i.e., soft tissue) portions. A series of experiments is conducted to understand the role of a general-purpose image encoder and its application in quantifying change in tumour burden during neoadjuvant chemotherapy in HGSOC. This work is the first to demonstrate the feasibility of a self-supervised image registration approach in quantifying NACT-induced localised tumour changes across the whole disease burden of patients with complex multi-site HGSOC, which could be used as a potential marker for ovarian cancer patient's long-term pathological response and survival.

Figures (5)

• Figure 1: Overview of the treatment schedule and characteristics of an advanced-stage high-grade serous carcinoma. Initial staging and diagnosis were based on CE-CT imaging and biopsy. Surgery was performed after two cycles of NACT. According to RECIST 1.1 criteria, the patient was classified as a partial responder. Pre- and post-treatment CE-CT scans revealed the metastatic nature of the disease, identifying 16 sites of disease burden with varying numbers of connected components (CC) and highlighting heterogeneity in lesion response.
• Figure 2: Image registration follows a multi-step, multi-resolution approach trained with a two-stage process (A) via the Two Step (C) and the Downsample (D) operators. The basic component $\Psi$ of the registration network is represented by a U-Net instance, taking as input two images and returning a displacement field (B). The downsample operator predicts the warp between two high-resolution images using a network that operates on low-resolution images, and the two-step operator predicts the warp between two images in two steps, first capturing the coarse transform via $\Phi$ and then the residual transform via $\Psi$.
• Figure 3: (A) Main characteristics of the patient cohort. (B) Sites of primary and metastatic HGSOC disease. (C) Distribution of tumour volumes by disease site.
• Figure 4: Baseline, follow-up, DVF, and Jacobian maps in axial view for HGSOC patients classified as a responder (top row) and a non-responder (bottom row) according to RECIST 1.1. Tumour segmentations are overlaid on top of both the DVFs and the Jacobian maps, showing longitudinal volume changes. Red and blue colours in Jacobian maps represent expansion and shrinkage respectively, whereas white colour shows volume preservation.
• Figure 5: (A) 3D and axial CT reconstructions showing segmented ROIs: hyper-dense, hypo-dense and soft tissue components relative to the whole omental tumour. (B) Total percentage of each tissue type for two disease sites at baseline and follow-up for the cohort.