Prostate cancer (PCa) is the second most prevalent cancer and the fifth leading cause of death in men worldwide [1]. While PCa responds well to treatment, bone metastases (BM) status represents a critical juncture in both medicine and prognosis, contributing to poor predictions and heightened mortality rates among patients with advanced disease [2]. Upon occurrence, BM leads to a range of skeletal complications, including hypercalcemia, pathological fractures, spinal compression, and bone pain, posing a significant challenge for doctors [3]. The BM state significantly influences prognosis and treatment approaches for PCa patients. Regrettably, screening for BM state in patients with Prostate-Specific Antigen (PSA) levels above 10 ng/ml using the European Urological Association (EUA) guidelines yields a mere 7% positive rate. Additionally, the prediction model's area under the receiver operating characteristic curve (AUC) was 0.79, underscoring the inadequacy of solely depending on clinical risk features to predict BM [4]. Thus, developing novel predictive models tailored to patients with substantial individual variations is crucial.

In 2012, Lambin [5], a Dutch scientist, introduced Radiomics—a method harnessing texture data from CT and MRI scans—which has since demonstrated its potential in discerning benign from malignant prostate tumors and refining prognosis evaluations [[6], [7], [8], [9], [10]]. Wang and Zhou et al. [11,12] each employed logistic regression approach to construct two different radiomics models to predict the BM state, excluding consideration of machine learning methodologies. Meantime, the model mandates dynamic contrast-enhanced (DCE) sequence images, which aren't suited for current biparametric magnetic resonance imaging (bpMRI) scanning strategies, limiting its broader applicability. Furthermore, Wang's model did not consider important variables such as Gleason score (GS) and International Society for Urological Pathology (ISUP) classification [12].

A convenient and accurate diagnosis of BM in PCa patients remains a significant challenge. In medical research, machine learning outperforms traditional algorithms in handling high-dimensional data. This study aims to compare different modeling techniques and determine the most effective radiomics-based models for predicting BM in newly diagnosed PCa patients.