Multiple myeloma (MM) is an incurable malignancy characterized by clonal plasma cell proliferation in the bone marrow and excessive monoclonal immunoglobulin production (1). MM constitutes 1.8% of all cancer cases, with an annual incidence rate of 4.6 to 6 cases per 10,000 individuals, thereby ranking as the second most prevalent hematologic malignancy (2, 3). In the past two decades, the overall survival rate for MM patients has significantly improved due to the introduction of novel therapeutic agents (4, 5). Concurrently, the development of sensitive methods for measuring deep therapeutic responses has led to the concept that sustained minimal residual disease (MRD) negativity is associated with remarkably low progression rates (6–8). However, despite these unprecedented treatment responses, 30-50% of patients fail to achieve or maintain MRD negativity, ultimately progressing to advanced stages of the disease (4). Moreover, in clinical practice, significant disparities in survival outcomes are observed among patients presenting with the same disease stages (9). Current prognostic staging systems have limitations in accurately stratifying risk, particularly for intermediate-stage disease (10, 11). Understanding the therapeutic effects and prognostic characteristics of patients with MM is essential for optimizing treatment strategies.

Despite advances in the treatment of MM, relapse remains common. Furthermore, there are no standard treatments for patients with advanced MM following primary therapies; complete responses are rare, with a median progression-free survival of 3–4 months and an overall survival of 8–9 months (10). Current assessment methods are often invasive, time-consuming, and difficult for some patients to undergo. Due to the wide variability in survival among MM patients, developing a more accurate and comprehensive therapeutic assessment system remains a major focus and challenge in clinical practice.

Accurate therapeutic effects and prognostic stratification are essential for selecting appropriate treatment strategies for MM, so as to prevent overtreatment and reduce the medical and financial burden. The International Myeloma Working Group recommends 18F-FDG PET/CT as one of the preferred imaging modalities for evaluating MM and other plasma cell disorders (3, 11). PET/CT integrates both anatomical and metabolic information, providing high sensitivity and specificity for assessing bone damage and detecting extramedullary disease (3). It can identify additional lesions in up to 40% of patients with early-stage MM (12). Compared with MRI and X-ray, PET/CT has superior capabilities for detecting extramedullary disease, making it an indispensable tool in the diagnostic and therapeutic assessment of MM (13). Many studies have indicated that imaging features such as focal lesions, metabolic activity, tumor volume, and extramedullary disease are associated with survival outcomes (14) However, most studies focus solely on imaging features, overlooking the integration of clinical, laboratory, and cytogenetic data for a comprehensive assessment. Moreover, current evaluations still largely depend on physicians’ visual interpretation and basic quantitative metrics such as lesion count, SUVmax, metabolic tumor volume, etc., which are subject to interobserver variability and inconsistent definitions. This underscores the need for more robust, reproducible, and integrative assessment methods.

Radiomics has emerged as an important field involving the automated or semi-automated extraction of quantitative features, including texture, intensity, and density, from medical imaging (15). Numerous studies have demonstrated the efficacy of radiomics in the differential diagnosis of cancer, the prediction of treatment response, and the prognosis of disease progression (16–18). In our prior studies, we successfully employed radiomics models based on MRI to differentiate between spinal MM and metastases (19, 20). Although radiomics studies based on PET/CT are gaining attention, their application in assessing treatment response for MM remains limited, mainly focusing on predicting overall survival or recurrence (21–26).

This study aimed to investigate the feasibility of utilizing radiomics based on 18F-FDG PET/CT imaging to evaluate the therapeutic efficacy in patients newly diagnosed with MM. Additionally, the research sought to develop a model integrating clinical and radiomic data to differentiate between MM patients exhibiting deep response and those with non-deep response to treatment.

The precise prediction of MM patients who achieve deep responses holds considerable importance for directing treatment, monitoring disease progression, evaluating prognosis, and enhancing the quality of life of patients. In this investigation, we developed radiomics models that leveraged PET/CT features to assess therapeutic responses in individuals newly diagnosed with MM. The principal findings indicated that the radiomics model, which incorporated PET/CT features, significantly surpassed models based solely on clinical or PET characteristics. Furthermore, the amalgamation of multimodal data within the PET + CT and PET + CT + Clinical models yielded optimal predictive performance, achieving area under the curve (AUC) values of 0.809 and 0.813, respectively. These findings highlight the advantages of integrating multimodal data to enhance prediction accuracy.

In this study, we examined the influence of clinical characteristics on the prediction of patients with DR in MM. Surprisingly, only gender emerged as a significant predictor of therapeutic efficacy following multivariate logistic regression analysis, with a notably better prognosis for females. Indeed, there exists controversy regarding the impact of gender on the prognosis of MM. The majority of studies suggest that female patients with MM generally experience better prognoses compared to their male counterparts (30–33), which aligns with our observations. Potential explanations may include healthier attitudes and behaviors among women, resulting in greater participation in health-promoting activities post-treatment (34). Additionally, men often present with more comorbidities at the time of MM diagnosis, which could adversely affect survival rates. Biological factors may also play a role in these gender disparities in MM outcomes. For instance, hormone-related pharmacokinetic variations in lymphoma indicate that elevated rituximab serum levels in females correlate with improved progression-free survival (35). Moreover, a greater presence of regulatory T cells (Tregs) in males may contribute to their less favorable outcomes (36). Nonetheless, other studies have reported no significant influence of gender on prognosis (36). Our findings emphasize the necessity of accounting for gender differences in the management of MM and suggest that future research should delve deeper into the underlying biological and behavioral mechanisms to refine personalized treatment strategies.

To date, both ISS and RISS are widely utilized for risk stratification in MM. However, due to the heterogeneity of this disease, a singular system may not adequately capture the nuances applicable to all patients. For those with non-secretory or low-secretory MM, where the tumor burden is elevated alongside low serum 2-MG levels, ISS staging may prove inadequate. Additionally, certain patients displaying low serum β2-MG levels during early disease stages may present with chromosomal translocation t(4;14) and other cytogenetic anomalies associated with poor prognosis. Zhou H et al. demonstrated in their study that both RISS and ISS failed to significantly differentiate early relapse risk (37). Our multivariate analysis revealed that neither ISS nor RISS effectively distinguished patients with deep responses, indicating that conventional staging systems may not accurately reflect the intricate biological complexities of MM. A more holistic approach is warranted for evaluating tumor burden in patients diagnosed with MM.

Radiomics is an emerging and rapidly advancing domain within the realm of medical imaging, focusing on the analysis of medical images to extract high-dimensional quantitative data that unveils concealed information not readily apparent to the naked eye (15, 17). A range of medical imaging modalities can facilitate radiomic analysis in patients with MM, with MRI being the most extensively utilized, followed by CT and PET (38). In comparison to MRI and CT, PET provides a more insightful approach for assessing the metabolic activity of MM lesions, thereby offering distinct advantages in evaluating patient responses to therapy. This rationale underpins the IMWG recommendation of 18F-FDG PET/CT as the current “gold standard” for assessing and monitoring responses to anti-myeloma treatment (1, 2). However, differentiating between focal and diffuse patterns remains a considerable challenge. Consequently, several studies have turned their focus to radiomics methodologies, illustrating the potential applications of radiomics derived from PET/CT in differential diagnosis, MRD detection, and prognosis prediction for MM (21). Reports on the application of radiomics in predicting therapeutic efficacy for patients with newly diagnosed MM are notably scarce. In our investigation, models predicated on radiomic features demonstrated a substantial advantage over clinical models in identifying newly diagnosed patients with deep responses, exhibiting significantly higher values of AUC, sensitivity, specificity, accuracy, and F1 score (Table 3). This superiority is likely attributable to radiomics’ capacity to capture intricate details of tumor heterogeneity, offering insights that transcend traditional clinical features. Given the high heterogeneity of MM, quantitatively characterizing its inter- and intra-tumoral variations could significantly enhance prognostic assessments. Importantly, our findings indicate that the integration of multiple data modalities yielded the most robust predictive performance, as exemplified by the PET+CT and PET+CT+Clinical models. These models achieved superior accuracy in distinguishing between patients with deep responses and those without, suggesting their potential utility for risk stratification in MM. While our results do not directly evaluate treatment outcomes, they imply that such stratification may inform personalized therapeutic strategies in the future. For instance, patients exhibiting a deep response might be spared from unnecessary intensive treatment, whereas those without could benefit from closer monitoring and timely adjustments to their therapy. Furthermore, this study underscored prospective trends in merging radiomics with other data types, with the goal of comprehensively elucidating MM through the unveiling of complex connections between imaging phenotypes and the molecular mechanisms driving disease progression (38).

At present, the commonly used first-line regimens may include bortezomib, lenalidomide and dasatuzumab. For patients with poor efficacy after 2 courses of treatment, a new chemotherapy combination can be selected according to the patient’s drug resistance. For lenalidomide-resistant patients, proteasome inhibitors bortezomib, ixazomib or a new generation of proteasome inhibitors carfilzomib, a new generation of immunomodulators pomalidomide and CD38 monoclonal antibody-based regimens can be used. For bortezomib-resistant patients, regimens based on carfilzomib, pomalidomide, and CD38 monoclonal antibodies can be used. For patients with triple drug resistance, a regimen consisting of carfilzomib, pomalidomide, and celinesol can be used. If the patient’s age and physical condition allow, autologous hematopoietic stem cell transplantation can also be considered (39–42).

This study does possess certain limitations that warrant attention. First, the sample size is relatively small, necessitating validation through larger, multi-center datasets to ascertain the generalizability of our findings. Second, although cytogenetic testing was conducted for all patients, as our study was explicitly designed to concentrate on radiomics-based prediction. The prognostic significance of genetic alterations will be explored in a subsequent investigation. Third, the methodology was reliant on traditional radiomics techniques involving manual or semi-automatic feature extraction, which may introduce a degree of subjectivity. Future endeavors will investigate the integration of deep learning methodologies to facilitate automated feature learning, potentially enhancing predictive accuracy and thereby broadening the clinical applicability of radiomics in MM.

In conclusion, the radiomics model derived from FDG-PET/CT exhibits considerable promise for clinical implementation in forecasting profound treatment responses in patients newly diagnosed with MM. By non-invasively capturing tumor heterogeneity and metabolic attributes, this model serves as an objective and reproducible instrument for stratifying patients based on their probability of attaining a deep response. The amalgamation of radiomic characteristics with pivotal clinical factors, such as sex, further bolsters predictive precision, underscoring its applicability in personalized therapeutic strategies.

This model has the potential to facilitate early treatment assessment, inform adjustments, prevent overtreatment in responders, and pinpoint patients who require closer surveillance. Future multicenter validation and the incorporation of genetic markers may enhance its clinical relevance in the realm of precision medicine for MM.