Noninvasive prediction of Ki67 proliferation index in breast cancer based on integrated photoacoustic imaging

BACKGROUND: Predicting Ki67 expression is crucial for understanding tumor proliferation and guiding personalized breast cancer treatment. Non-invasive methods remain limited, underscoring the need for innovative imaging-based approaches to enhance molecular subtyping and clinical decisions. PURPOSE: This study aimed to develop a predictive model integrating photoacoustic/ultrasound (PA/US) imaging data and clinical variables to differentiate high and low Ki67 expression levels in breast cancer. It sought to identify imaging and clinical factors associated with Ki67 expression, contributing to the molecular subtyping of breast cancer. METHODS AND MATERIALS: In this study, 336 breast tumors were analyzed and divided into high Ki67 expression (≥14%) and low Ki67 expression (<14%) groups. The samples were randomly split into training and test sets at a 7:3 ratio. Statistical methods included t‑tests and rank‑sum tests, with independent predictors identified through univariate and multivariate logistic regression analyses. Four predictive models were developed: Model A (clinical factors), Model B (clinical factors combined with ultrasound features), Model C (combining clinical, ultrasound, and photoacoustic oxygen saturation [PA‑SO₂]), and Model D (clinical factors combined with PA‑SO₂). RESULTS: Using univariate and multivariate logistic regression analysis, four independent predictive factors were identified: histological grade, axillary lymph node status (ALN), intratumoral color Doppler flow imaging (Inter CDFI), and PA‑SO₂. Based on these factors, four logistic regression models were constructed for predicting high vs. low Ki67 expression: Model A (clinical factors only): histological grade + ALN; Model B (clinical + ultrasound): Model A + Inter CDFI; Model C (comprehensive model): Model B + PA‑SO₂; Model D (comprehensive model): Model A + PA‑SO₂. In the test set, the areas under the receiver operating characteristic curve (AUCs) with 95% confidence intervals were as follows: Model A, 0.781 (0.696-0.866); Model B, 0.779 (0.691-0.867); Model C, 0.823 (0.739-0.908), and Model D, 0.807 (0.721-0.892). Model C demonstrated the highest diagnostic efficiency for distinguishing Ki67 expression levels. CONCLUSION: This study developed a predictive model incorporating histological grade, ALN, Inter CDFI, and PA‑SO₂ to estimate Ki67 expression levels in breast cancer. This model provides a valuable tool for early prognosis, aiding molecular classification and facilitating the prompt initiation of personalized treatment strategies.