IVIM-based habitat imaging for non-invasive assessment of Ki-67 proliferation status in cervical cancer
Year: 2026
Type: article
Study design: Observational Study
Abstract: Objective This study aimed to evaluate whether voxel-level IVIM-based habitat analysis supports exploratory pretreatment stratification of Ki-67 proliferation status in cervical cancer. Materials and methods Between January 2019 and June 2025, medical records of 149 patients with histopathologically confirmed cervical cancer were initially screened. Before image-quality assessment, 49 were excluded because IVIM MRI had been performed after chemoradiotherapy (n = 5), cervical biopsy sampling (n = 21), or lesion excision (n = 10), or because no pathological Ki-67 immunohistochemistry result was available (n = 13). Of the remaining 100 patients, 17 were excluded because of severe image artefacts (n = 11) or lesions too small for reliable ROI delineation (n = 6), leaving 83 patients for analysis. Patients were stratified into high Ki-67 (Ki-67 >70%, n = 38) and low/equal Ki-67 (Ki-67 ≤70%, n = 45) groups. Solid-tumour regions were analysed with patient-level K-means clustering using normalised D, f, and ADC features; D* was excluded from clustering because of voxel-level fitting instability. Habitat labels were harmonised according to quantitative parameter profiles and treated as image-defined phenotypes. Habitat-derived features, conventional ROI parameters, and logistic models were evaluated for Ki-67 stratification. Results Hab1 ADC was lower in the high- than in the low/equal-Ki-67 group (median [Q1–Q3], 0.947 [0.862–1.161] vs. 1.257 [1.113–1.455] × 10 −3 mm 2 /s; P < 0.001; BH q = 2.04×10 −6 ), whereas Hab3 volume proportion was higher (0.396 [0.341–0.488] vs. 0.279 [0.190–0.390]; P = 0.002; BH q = 0.006). Hab1 ADC outperformed conventional ROI ADC for identifying high Ki-67 status [AUC, 0.826 (95% CI, 0.726–0.909) vs. 0.563 (95% CI, 0.429–0.695); paired DeLong P < 0.001]. The combined Hab1 ADC and Hab3 volume proportion model showed the numerically highest discrimination (apparent AUC, 0.877; 95% CI, 0.794–0.938), but this numerical improvement over Hab1 ADC alone was not statistically significant (paired DeLong P = 0.056). Fixed-feature leave-one-out validation of the combined model yielded an AUC of 0.855 (95% CI, 0.771–0.923) and a Brier score of 0.157 (95% CI, 0.112–0.204). Conclusion IVIM-based habitat analysis showed exploratory potential for non-invasive Ki-67 stratification by isolating a low-diffusivity imaging compartment otherwise obscured by conventional whole-tumour averaging. The combined habitat model should be interpreted as a research-stage imaging biomarker, not a clinical decision tool, and requires prospective external validation before clinical translation.
Source: Frontiers in Oncology
Language: English
Topic: MRI in cancer diagnosis
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,801
OpenAlex ID: W7221768235