Establishment and validation of novel predictive models to predict bone metastasis in newly diagnosed prostate adenocarcinoma based on single-photon emission computed tomography radiomics.

Establishment and validation of novel predictive models to predict bone metastasis in newly diagnosed prostate adenocarcinoma based on single-photon emission computed tomography radiomics. | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Establishment and validation of novel predictive models to predict bone metastasis in newly diagnosed prostate adenocarcinoma based on single-photon emission computed tomography radiomics. ning wang, shihui qu, weiwei kong, qian hua, zhihui hong, zengli liu, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3869436/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 4 You are reading this latest preprint version Abstract Purpose In order to establish and validate novel predictive models for predicting bone metastasis (BM) in newly diagnosed prostate adenocarcinoma (PCa) in single-photon emission computed tomography radiomics. Method In a retrospective review of clinical SPECT database, 176 patients (training set: n = 140; validation set: n = 36) who underwent SPECT/CT imaging and histologically confirmed with newly diagnosed PCa from June 2016 to June 2022 were enrolled. Radiomic features were extracted from ROI in a targeted lesion of each patient. Clinical features, including age, t-PSA, and Gleason grades, were included. Statistical tests were then used to eliminate irrelevant and redundant features. Finally, three types of optimized models were constructed for the prediction. Furthermore, 5-fold cross-validation was applied to obtain the sensitivity, specificity, accuracy, and area under the curve (AUC) for performance evaluation. The clinical usefulness of the multivariate models was estimated through decision curve analysis (DCA). Results Radiomics signature consisting of 27 selected features was significantly correlated with bone status(P < 0.01 for both training and validation sets). Collectively, the models showed good predictive efficiency. The AUC values ranged from 0.87 to 0.98 in four models. The AUC value of the human experts was 0.655 and 0.872 in the training and validation groups, respectively. Most radiomic models showed better diagnostic accuracy than human experts in the training group and the validation group. DCA also demonstrated the superiority of the radiomics models compared to human experts. Conclusion Our proposed models, which incorporate SPECT-based radiomics signature and clinical risk factors, could be a promising auxiliary means to assist radiologists or medical physicians in their subsequent workup to confirm the diagnosis of BM. prostate adenocarcinoma radiomics single-photon emission computed tomography bone metastasis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction In 2020, prostate cancer (PCa) was the second most common male malignancy around the world, with potentially devastating effects on the male genitourinary system [ 1 , 2 ] . -oOver 90% of patients with early-stage cancer survive 5 years after diagnosis; nevertheless, once metastasis occurs, the chances of survival drop dramatically to < 40% within 5 years [ 3 ] , as bone lesions are rarely eradicated completely. Notably, PCa is the only solid tumor where bone metastasis (BM) precede visceral metastases. Patients with BMs survival rates are < 10% after 5 years. Common sites for bone metastases should be monitored in patients with PCa to maximize the effects of treatment [ 4 , 5 ] . Because of its invasive nature and high procedural risks (such as damage to the vertebral artery or spinal cord), bone biopsies are rarely used for clinical diagnosis and treatment; however, they are the gold-standard for identifying benign and malignant lesions. There is an urgent and unmet need for a non-invasive method of distinguishing benign from malignant bone lesions. In patients with bone lesions, 99m Tc-labeled methylene diphosphonate ( 99m Tc-MDP) whole-body scans have been used, but their specificity was low [ 6 , 7 ] . Despite undergoing imaging tests such as computed tomography (CT), magnetic resonance imaging (MRI), or bone scans, disease recurrence may be diagnosed late due to the low sensitivity of these techniques [ 8 ] . Single-photon emission computed tomography/computed tomography (SPECT/CT) combines anatomical and metabolic functions to improve anatomic localization and bone imaging specificity [ 9 , 10 ] . The imaging characteristics of bone metastases and benign bone lesions were similar, according to several research studies [ 11 ] . After SPECT/CT examinations, 14.3% of patients still had an equivocal diagnosis. It was difficult to distinguish between bone metastases and benign bone lesions [ 12 ] . Additionally, there was no way to quantify the intensity, uniformity, or heterogeneity of lesion distribution, so SPECT/CT diagnosis depends on physicians' experience, which was always subjective [ 13 ] . Healthcare is being transformed by artificial intelligence (AI), cognitive tasks performed by computers and based on available data. AI can be used to identify unique or complex imaging features and facilitates quantitative evaluation [ 14 ] . In machine learning (ML), a subfield of AI, an algorithm is used to mathematically analyze the distribution of pixels (voxels) in medical gray-scale images, which can extract a large number of features from medical images in high throughput. Automatic or semi-automatic analysis methods are used to convert imaging data into a mining data space to quantify the heterogeneity of lesions. It has good application in differential diagnosis, histological classification, treatment options and prognosis analysis of diseases. There is growing interest in the use of AI to aid physicians who assess lesions. AI can surpass subjective visual interpretation in order to obtain additional information about tumor behavior and pathophysiology that is otherwise not inferable by the human eye and commonly used technologies. Medical professionals who visually evaluate a large number of uncertain images will unavoidably make mistakes. In order to improve the survival rate of patients with PCa, accurate predictive models are needed. Radiomics models derived from SPECT/CT images can effectively discriminate between vertebral bone metastases and benign bone disease [ 15 ] . Unfortunately, only a few studies have used SPECT/CT radiomics to study bone metastases associated with PCa. We investigated the feasibility of SPECT/CT imaging-based radiomics for identifying and diagnosing bone metastases from benign lesions in patients with PCa, mining more accurate digital information from the target area of SPECT images, and seeking more accurate guidance for patient treatment. Materials and Methods 2.1. Population selection We retrospectively analyzed a dataset of SPECT/CT images from 176 patients with biopsy-proven prostatic carcinoma between June 2016 and June 2022. This retrospective study was approved by the institutional review board, and the requirement for informed consent was waived. The patients were grouped randomly into the training (n = 140) and validation (n = 36) cohorts, with a ratio of 8:2. The major criteria for inclusion in the study were as follows: a)Eligible patients with histologically on-set confirmed PCa) All patients received a pre-treatment workup with imaging studies; c)clinical and imaging data were complete and can be followed up by telephone. We excluded: 1) patients with a history of prostate surgery, chemoradiotherapy, hormone therapy, and targeted therapy; 2) Patients with a history of malignant tumors; 3)images were unable to be evaluated owing to unsatisfactory quality; 4) region of interest (ROI) cannot be outlined clearly and correctly. Details concerning the flow of studies through the selection process are shown in Fig. 1 . Baseline clinicopathologic data, including age, prostate-specific antigen (PSA) level, Gleason score (GS), and histological type, were derived from electronic medical records. Bone status, including the presence or absence of BMs, was clearly determined by later follow-up outcome and MR imaging data. ECT for bone metastases was characterized by multiple radionuclide concentration areas [ 14 ] , we select 1–3 typical lesions (mainly in axial bones, limb bones, pelvis, etc.) from multiple lesions to extract features [16]. Benign lesions are also selected radionuclide concentration areas, mainly in the limb bones, ribs, pelvis, according to imaging follow-up and clinical progress. Laboratory analysis of PSA was performed via routine blood tests within 1 week before treatment initiation or ultrasound-guided biopsy. Total serum PSA level thresholds were ≤ 4 ng/mL, 4 ~ 100 ng/mL, and > 100 ng/mL, which was determined using radioimmunoassay methods for determining total PSA by Quest Diagnostics and in accordance with the normal range used at our institution. 2.2. Image acquisition and segmentation All the acquisition procedures were completed on a 16­section SPECT/CT scanner (dual head gamma camera Symbia Intevo bold, Siemens, Erlangen, Germany). The WBS was acquired within 2h after intravenous administration of 15–25 mCi 99m Tc-MDP (Shanghai Syncor Medicine Corp. Ltd. (Soochow, China)), then SPECT/CT was performed immediately for further diagnosis if a suspicious lesion was found on the WBS. With a 256×256 matrix by using a magnification of 1.00×, the SPECT images were acquired with a 15% window centered around a 140-keV photopeak. CT scanning parameters were as follows: tube voltage, 130 kV; automatic mAs control (reference mAs 120); rotation time, 0.6 seconds; pitch, 0.8; detector configuration, 1.0 mm; and beam width, 10 mm. Coronal and sagittal views were reconstructed using 2 mm thick slices. The CT reconstruction used a hard filter (B80s). Image reconstruction program was carried out in the SYngo workstation (Syncor Medicine Corp. Ltd., Soochow, China). The image fusion program was Symbia.net. 2.3. Image analysis and expert human qualitative classification After summarizing all available clinical data to determine the nature of the lesion, we concluded that the study’s diagnostic criteria were based on the comprehensive pathological biopsy, laboratory indicators, follow-up imaging, and clinical progression. All SPECT images were independently evaluated by two experienced readers (MBM and XYC) with more than 10 years of experience in qualitative classification and diagnosis. The lesions were visually classified in a double-blind manner, without access to pathological test results. The readers were only given the Total Prostate-Specific Antigen (t-PSA) and Gleason scores of the patient being read. The human expert diagnostic findings were evaluated using weighted Kappa statistics to obtain interobserver agreement. The main criteria for the human expert's qualitative classification of bone metastases were bone changes on SPECT images and abnormal uptake of 99 m Tc-MDP in the corresponding areas. 2.4.Lesion segmentation and feature extraction The patients’ SPECT images were imported into an open source three-dimensional (3D) slicer V4.13.0 ( https://www.slicer.org/ ) software in DICOM format. Without knowing the pathological results, the region of abnormal 99mTc-MDP uptake on each image was described as a region of interest (ROI) by two radiologists with more than 10 years of work experience. The radiologists carefully identified the edge of the lesion and delineated along the edge of the lesion in SPECT images. And they are distributed in different parts. To assess interobserver reliability for ROI delineation, 30 lesions were randomly selected for secondary outlining [ 16 ] . Before extracting the target features, the images were normalized using µ ± 3σ (where µ is the average value of the image gray value and σ is the SD of the image gray value) to reduce the impact of brightness and contrast on the images’ gray value. After segmenting the images, we used PyRadiomics [version 3.0.1, van Griethuysen, J. J. M., Fedorov, A., Parmar, C., Hosny, A., Aucoin, N., Narayan, V., Beets-Tan, R. G. H., Fillon-Robin, J. C., Pieper, S., Aerts, H. J. W. L. (2017). Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Research, 77(21), e104–e107] to analyze the ROI. We divided candidate features and the radiomics parameters into the following categories: First Order Statistics, Shape-based, Gray Level Cooccurence Matrix, Gray Level Run Length Matrix, Gray Level Size Zone Matrix, Neigbouring, Gray Tone Difference Matrix, Gray Level Dependence Matrix and wavelet. Detailed information about radiomics features was presented previously [ 17 ] . 2.5. Radiomics feature Selection and ML Model Establishment We used the Minimum-Redundancy Maximum-Relevance (mRMR)algorithm to determine which radiomic features were associated with differential diagnosis between benign and malignant lesion [ 18 ] . The mRMR algorithm identifies features that are highly correlated with bone status, but minimally correlated with other features in order to reduce overfitting of the model. The LASSO method is suitable for the regression of high-dimensional data and can be used to select meaningful features using the non-zero coefficient. This process included a 5–fold cross-validation scheme to minimize selection bias. When constructing the model, features’ importance was determined by their selection probabilities. Image features with a selected frequency > 90% were used to re-fit the final imaging model. Finally, four supervised ML algorithms were implemented in this study: LogisticRegression, RandomForest, Xgboost, and CatBoost. In order to obtain high accurate models, we assessed model performance using receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) in the training and validation cohorts, respectively. Figure 4 summarizes the models’ ROC curves. 2.6 Statistical analysis IBM SPSS (version 25.0) software was used for the statistical analyses, and p < 0.05 was considered statistically significant. For determining between-group differences for categorical variables, we used the chi-square test for two independent samples; the Mann-Whitney U test for two independent datasets was used for continuous variables. Weighted Kappa statistics were used to evaluate interobserver agreement. We used AUC and ROC curves to evaluate the predictive abilities of radiomic and image features. The DeLong test was used to detect statistical differences among the five predictive models. Results 3.1 Patients’ clinicopathologic features Figure 2 is a detailed study flow chart. We enrolled 176 patients with PCa, including 81 patients with (mean age 75.27 ± 7.85 years) and 95 patients without BM (mean age 73.20 ± 9.11 years). Of these, 140 patients were assigned to the training set, and 36 to the validation set. There were no statistically significant differences in age ( p = 0.11) or t-PSA distribution ( p = 0.137) between the training (n = 140) and validation (n = 36) cohorts. Gleason grade and histological type were significantly related to BMs in both groups ( p < 0.05). The patients’ clinical characteristics are included in Table 1 . Table 1 Risk factors for bone metastasis in the training and validation cohorts BM(+)1 None-BM(-)2 P Age,mean ± SD,years 75.27 ± 7.85 73.20 ± 9.1 0.11 Gleason score(biopsy) < 0.001 GS(≤ 7)1 7(8.53) 32(34.41) GS(4 + 4 = 8)2 35(42.68) 30(32.56) GS(9–10)3 40(48.78) 30(32.56) NA 1(1.08) t-PSA level(ng/ml) 100 49(60.49) 15(15.79) NA 11(11.58) Histological type < 0.001 Adenocarcinoma 28(34.57) 18(18.95) Acinar adenocarcinoma 52(64.20) 75(78.95) NA 1(1.23) 2(2.11) NOTE.P value is derived from the univariable association analyses between each of the clinicopahologic variables and Bone status. *P value < 0.05 was considered to be statistically significant. 3.2 Construction of radiomics tags After removal by high correlation, we extracted 218 features from each ROI. We then constructed four different models based on corresponding features using LASSO; the models were compared using the AUC values, as shown in Figs. 3 B and C. According to the weighting coefficient corresponding to the feature (Fig. 3 A), a radiomics formula was obtained and used to calculate the histological score for each lesion in the training and verification groups. The radiomic features and corresponding coefficients are included in Supplementary. 3.3. Performance of different prediction models The four radiomic models in the training cohort were constructed based on different factors, with AUC values ranging from 0.87 to 0.98. The weighted Kappa value was 0.611, indicating good interobserver agreement. The AUC values of qualitative classification by human experts were 0.655 and 0.872 in the training and validation cohorts, respectively. Table 2 includes information on the models’ accuracy and the ROC curves are included in Fig. 4 . Logistic Regression showed the highest AUC values in the training and test groups (0.984, 0.916).The DeLong test was used for pairwise comparisons of the five prediction models to determine if any between-model differences in AUC were statistically significant. Significant differences were observed in the AUC between human experts' qualitative classification and the others ( p 0.05). We used the DCA curve to verify the models’ net benefit for predicting BMs. When the decision curve threshold was set at 0–1, the predictive models always outperformed the human experts. (Of note, the Logistic Regression model had a slightly higher clinical gain than the other models.) The decision curve was shown in Fig. 5 . Table 2 The diagnostic ability of each model for discriminating Bone metastasis of prostate cancer from benign bone lesions Model The training cohort The validation cohort AUC Specificity Sensitivity Accuracy PPV NPV AUC Specificity Sensitivity Accuracy PPV NPV Logistic Regression 0.97 0.96 0.89 0.93 0.95 0.91 0.98 1 0.88 0.94 1 0.9 Random Forest 0.99 0.98 0.95 0.97 0.98 0.96 0.91 0.95 0.82 0.89 0.93 0.86 Xgboost 0.99 0.97 0.92 0.95 0.97 0.94 0.91 0.89 0.82 0.86 0.88 0.85 CatBoost 0.99 1 0.98 0.99 1 0.99 0.93 0.95 0.82 0.89 0.93 0.86 Human experts 0.66 0.67 0.80 0.66 0.74 0.74 0.87 0.65 0.79 0.78 0.73 0.71 AUC,the area under the ROC curve; PPV ,postive predictive value; NPV ,negative predictive value. The models’ curves were nearly ideal, indicating that the models had superior adaptability and predictive ability. The calibration curve was shown in Fig. 6 . Collectively, the predictive models were accurate and outperformed human experts. Importantly, the Logistic Regression model provided the optimal prediction. Discussion Given the emerging focus on patient-centered therapies, radiomics can help oncologists bridge the gap between histological findings and real microenvironment heterogeneity. Radiomics may be able to mitigate the limitations of subjective imaging-based evaluations and provide objective assessments of cancer heterogeneity while quantifying patients’ survival. We built and validated a radiomics and human expert model able to distinguish BMs from normal bone in patients with PCa based on SPECT/CT images. The ML model achieved excellent binary classification performance in patients with PCa. This was one of the first SPECT imaging studies of PCa bone metastases. We also created five prediction models, rather than the 1–3 models produced by previous studies. We selected a whole-body scan strategy to extract features that minimized complexity while achieving effective radiomics. With the help of image characterization algorithms, radiomic features can be quantified as criteria for determining tumor heterogeneity [ 19 ] . Our proposed models provide are easy-to-use, quantitative, and individualized tools for predicting BMs. Use of our models may help avoid inappropriate surgical procedures for occult BM-positive patients while reducing the risk of BMs for high-risk patients. We used LASSO to reduce the regression coefficient and construct the radiomic signature. Our final models included the 27 best potential predictors of BMs. Importantly, the radiomics signature was a better predictor of BMs than independent clinical risk factors. This suggests that, compared with gold-standard methods, the radiomics method was better at predicting BM risk. A prior multi-factor regression analysis revealed that radiomic features and GS independently predicted bone volume in patients with PCa. The GS is used to evaluate tumor aggressiveness [ 20 ] . The GS is used to assess risk in patients with PCa. The score could also be used to guide ML interpretation of SPECT images obtained from patients with BMs. Meanwhile, t-PSA may be a more effective predictor of PCa. However, the correlation between t-PSA and BM was not significant in this study. Several factors may affect the accuracy of biopsy t-PSA levels, including inappropriate handling prior to blood collection, hormone therapy, and inflammation. Radiomics seeks to develop clinical models to identify and assess patients' condition and streamline clinical management. Most prior studies used CT/MRI imaging features to establish BM prediction models. Of the studies that used radiomics for predicting bone disease, differentiating benign from malignant bone lesions, and characterizing malignant bone lesions, 85% were MRI-based and 15% were CT-based [ 21 ] . Wang et al. used multiparametric PCa MRI to extract radiomic features, which were combined with free PSA levels and Gleason scores to predict the presence/absence of BMs in patients with PCa [ 22 ] . The AUC of Wang’s model was 0.93 compared to 0.98 for our best model. Hinzpeter et al. proposed classifying patients' bones as either BM or normal bone using a training dataset. The researchers created a gradient-enhanced tree from 11 selected radiological features measured by quantitative methods from CT images [ 23 ] . The training model had a classification accuracy of 0.85 (95% confidence interval: 0.76—0.92, p < 0.001), a sensitivity of 78%, and a specificity of 93%. Several of our models’ strengths warrant mentioning. First, we used common SPECT features to assess BM in patients with PCa and provided clinicians with a quantitative, easy-to-use tool to predict BM risk. This effort culminated in the creation of four radiomic models. Despite this, the evidence supporting the prediction of BMs in patients with PCa using SPECT radiomic features was insufficient. A recent study by Jin et al. [ 15 ] evaluated a radiomics model based on CT and SPECT images to discriminate between vertebral BMs and benign bone disease. BMs that occur secondary to PCa cancer are well-known. Researchers continue to mine SPECT databases for features that distinguish benign and malignant bone tumors. However, the associated cases based on characterization and ML generally suffer from low accuracy. Our study model AUC range was 0.88–0.93, considered acceptable given the inherent characteristics of SPECT image textures (low resolution and image clarity). All ML models outperformed human experts for distinguishing BMs from benign bone in the training and validation sets. Compared to gold-standard methods of image assessment, radiomics is a novel tool for extracting image information using high-throughput methods [ 24 ] .The resultant information is used for disease typing and grading, gene localization, early treatment, and prognostication. Some studies have shown that radiomics performs better than traditional clinical methods for non-invasive classification and diagnosis of diseases [ 25 , 26 ] . Expanded use of SPECT may allow most patients with PCa to detect bone metastases early and evaluate treatment efficacy given that SPECT is more sensitive than MRI or PET/CT and widely available to patients. There is an obvious need for continued research in this area. Because our findings are only applicable to PCa, our results cannot be generalized to patients with other diseases. Future, well-powered studies are needed to validate use of our models with patients carrying other diagnoses. All lesions’ statuses were confirmed by biopsy or follow-up data; if the final diagnosis was uncertain, the lesion was simply removed. Although time-consuming, these exclusions ensure our study’s rigor. Finally, we excluded patients who received treatment because a certain percentage of patients experience flare-ups and osteogenic reactions after chemotherapy or radiotherapy, which may also affect the radiotracer uptake. Our study has several limitations that warrant consideration. This was a retrospective, and not prospective, research design. Our study cohort was relatively small; this was the most critical limitation of this study. We did not validate our findings using external data and there is most certainly a selection bias. Non-uptake areas on SPECT were considered normal bone. Lastly, the radiomic features we identified may not be generalizable to patients with other diseases. In the end, we were unable to perform a detailed histopathological analysis of each case; BMs and benign bone lesions were primarily confirmed based on pathological findings, radiographic follow-up, and disease progression. PCa radiomics should emerge as a consistent and objective tool for setting up clinical trials and tailoring treatment, as it allows for accurate assessment of patients at the time of diagnosis and during treatment. Conclusion Models developed using different algorithms and radiomics features performed better than human experts at distinguishing benign bone from BMs secondary to PCa. These findings suggest that AI models can be applied to clinical settings. The Logistic Regression model was best predicting BMs. Declarations Data Availability Statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Ethics Statement This study was conducted in accordance with the guidelines of the Declaration of Helsinki. The studies involving human participants were reviewed and approved by Ethics Committee of The Second Affiliated Hospital of Soochow University. The ethics committee waived the requirement of written informed consent for participation. Author Contributions Conception and design: NW and SQ. Collection and assembly of the data: WK. Development of the methodology: QH. Data analysis and interpretation: ZH. Manuscript writing: All authors. Manuscript review: YS and ZL. All authors contributed to the article and approved the submitted version. Conflict of Interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. References Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209–49. Culp MB, Soerjomataram I, Efstathiou JA, Bray F, Jemal A. Recent Global Patterns in Prostate Cancer Incidence and Mortality Rates. Eur Urol. 2020;77(1):38–52. Norum J, Nieder C. Treatments for Metastatic Prostate Cancer (mPC): A Review of Costing Evidence. Pharmacoeconomics 2017, 35 (12), 1223–36. Chaffer CL, Weinberg RA, Perspective A. on Cancer Cell Metastasis. Science 2011, 331 (6024), 1559–1564. Svensson E, Christiansen CF, Ulrichsen SP, Rorth MR, Sorensen HT. Survival after bone metastasis by primary cancer type: a Danish population-based cohort study. BMJ Open 2017, 7 (9), e016022. Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45(2):228–47. Chen Z, Chen X, Wang R. Application of SPECT and PET / CT with computer-aided diagnosis in bone metastasis of prostate cancer: a review. Cancer Imaging. 2022;22(1):18. Calais J, Cao M, Nickols NG. The Utility of PET/CT in the Planning of External Radiation Therapy for Prostate Cancer. J Nucl Med. 2018;59(4):557–67. Alqahtani MM, Fulton R, Constable C, Willowson KP, Kench PL. Diagnostic performance of whole-body SPECT/CT in bone metastasis detection using (99m)Tc-labelled diphosphate: a systematic review and meta-analysis. Clin Radiol 2020, 75 (12), 961 e11-961 e24. Langsteger W, Rezaee A, Pirich C, Beheshti M. (18)F-NaF-PET/CT and (99m)Tc-MDP Bone Scintigraphy in the Detection of Bone Metastases in Prostate Cancer. Semin Nucl Med. 2016;46(6):491–501. Abikhzer G, Srour S, Keidar Z, Bar-Shalom R, Kagna O, Israel O, et al. Added Value of SPECT/CT in the Evaluation of Benign Bone Diseases of the Appendicular Skeleton. Clin Nucl Med. 2016;41(4):e195–9. Guezennec C, Keromnes N, Robin P, Abgral R, Bourhis D, Querellou S et al. Incremental diagnostic utility of systematic double-bed SPECT/CT for bone scintigraphy in initial staging of cancer patients. Cancer Imaging 2017, 17 (1), 16. De Schepper S, Gnanasegaran G, Dickson JC, Van den Wyngaert T. Absolute Quantification in Diagnostic SPECT/CT: The Phantom Premise. Diagnostics 2021, 11 (12). Hosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts H. Artificial intelligence in radiology. Nat Rev Cancer 2018, 18 (8), 500–10. Jin Z, Zhang F, Wang Y, Tian A, Zhang J, Chen M, et al. Single-Photon Emission Computed Tomography/Computed Tomography Image-Based Radiomics for Discriminating Vertebral Bone Metastases From Benign Bone Lesions in Patients With Tumors. Front Med (Lausanne). 2021;8:792581. Xue C, Yuan J, Lo GG, Chang ATY, Poon DMC, Wong OL, et al. Radiomics feature reliability assessed by intraclass correlation coefficient: a systematic review. Quant Imaging Med Surg. 2021;11(10):4431–60. Zwanenburg A, Vallieres M, Abdalah MA, Aerts H, Andrearczyk V, Apte A, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020;295(2):328–38. Peng HC, Long FH, Ding C. Feature selection based on mutual information: Criteria of max-dependency, max-relevance, and min-redundancy. Ieee Trans Pattern Anal Mach Intell 2005, 27 (8), 1226–38. Limkin EJ, Sun R, Dercle L, Zacharaki EI, Robert C, Reuze S, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol. 2017;28(6):1191–206. Brunese L, Mercaldo F, Reginelli A, Santone A. Radiomics for Gleason Score Detection through Deep Learning. Sens (Basel). 2020;20:18. Faiella E, Santucci D, Calabrese A, Russo F, Vadala G, Zobel BB et al. Artificial Intelligence in Bone Metastases: An MRI and CT Imaging Review. Int J Environ Res Public Health 2022, 19 (3). Wang Y, Yu B, Zhong F, Guo Q, Li K, Hou Y, et al. MRI-based texture analysis of the primary tumor for pre-treatment prediction of bone metastases in prostate cancer. Magn Reson Imaging. 2019;60:76–84. Hinzpeter R, Baumann L, Guggenberger R, Huellner M, Alkadhi H, Baessler B. Radiomics for detecting prostate cancer bone metastases invisible in CT: a proof-of-concept study. Eur Radiol. 2022;32(3):1823–32. Tomaszewski MR, Gillies RJ. The Biological Meaning of Radiomic Features. Radiol 2021, 298 (3), 505–16. Lohmann P, Franceschi E, Vollmuth P, Dhermain F, Weller M, Preusser M, et al. Radiomics in neuro-oncological clinical trials. Lancet Digit Health. 2022;4(11):e841–9. Yang M, Hu P, Li M, Ding R, Wang Y, Pan S, et al. Computed Tomography-Based Radiomics in Predicting T Stage and Length of Esophageal Squamous Cell Carcinoma. Front Oncol. 2021;11:722961. Cite Share Download PDF Status: Under Review Version 1 posted Reviewers agreed at journal 17 Jan, 2024 Reviewers invited by journal 16 Jan, 2024 Editor assigned by journal 15 Jan, 2024 First submitted to journal 15 Jan, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3869436","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":267524380,"identity":"a8d76630-d17b-4640-b422-2d25eb1ae7c2","order_by":0,"name":"ning wang","email":"","orcid":"","institution":"Second Affiliated Hospital of Soochow University","correspondingAuthor":false,"prefix":"","firstName":"ning","middleName":"","lastName":"wang","suffix":""},{"id":267524381,"identity":"7776c3e0-efb1-4327-910a-fd089d20aa57","order_by":1,"name":"shihui qu","email":"","orcid":"","institution":"Second Affiliated Hospital of Soochow","correspondingAuthor":false,"prefix":"","firstName":"shihui","middleName":"","lastName":"qu","suffix":""},{"id":267524382,"identity":"45ca53e5-9b5f-402f-bebc-0a593e4c78f2","order_by":2,"name":"weiwei kong","email":"","orcid":"","institution":"Second Affiliated Hospital of Soochow University","correspondingAuthor":false,"prefix":"","firstName":"weiwei","middleName":"","lastName":"kong","suffix":""},{"id":267524383,"identity":"6f3acd04-f9e3-423b-9183-90e7e81e654e","order_by":3,"name":"qian hua","email":"","orcid":"","institution":"Second Affiliated Hospital of Soochow University","correspondingAuthor":false,"prefix":"","firstName":"qian","middleName":"","lastName":"hua","suffix":""},{"id":267524384,"identity":"a2e37b70-b76e-47a4-abb4-1d37acafcdbb","order_by":4,"name":"zhihui hong","email":"","orcid":"","institution":"Second Affiliated Hospital of Soochow University","correspondingAuthor":false,"prefix":"","firstName":"zhihui","middleName":"","lastName":"hong","suffix":""},{"id":267524385,"identity":"4655affe-ab20-4726-b522-3d8040d75886","order_by":5,"name":"zengli liu","email":"","orcid":"","institution":"Second Affiliated Hospital of Soochow University","correspondingAuthor":false,"prefix":"","firstName":"zengli","middleName":"","lastName":"liu","suffix":""},{"id":267524386,"identity":"61481df7-cf01-4588-9347-ca52e1862e95","order_by":6,"name":"yizhen shi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8UlEQVRIiWNgGAWjYDACCcYGhgSGAwwM7A0MDDwgkQNEa+E5QLQWBqgyiQQitcjPbm6TePDnjpy55PNnEm9qGOT4biQwfi7Ao8XgzsE2iQSeZ8aWs3PMJOccYzCWvJHALD0DnxaJRKAWicOJG27nsEnzsDEkbriRwMbMg89hM0BaDA7Xb7h5/Jk0zz+GeoJaGG6AtCQcTjC4wWAmzdvGAGQQ0GJwI7HZIuHAYcMNZ3KMLef2SRjOPPOwWRq/w9If3vzx57C8wfHjD2+8+WYjz3c8+eBnvA5jYGCRQOKA2MDIJQCYPxBSMQpGwSgYBSMcAAA7WlDqhYiOTwAAAABJRU5ErkJggg==","orcid":"","institution":"Second Affiliated Hospital of Soochow University","correspondingAuthor":true,"prefix":"","firstName":"yizhen","middleName":"","lastName":"shi","suffix":""}],"badges":[],"createdAt":"2024-01-16 10:10:35","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3869436/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3869436/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49838323,"identity":"7ed7b037-36c7-4f20-8b1a-4728a93b7383","added_by":"auto","created_at":"2024-01-18 19:57:28","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":116927,"visible":true,"origin":"","legend":"\u003cp\u003eRecruitment pathway for patients in this study.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/25134b5476975221764b1ab2.png"},{"id":49838328,"identity":"947ffd8e-c40e-45b2-860f-b62e16c0781f","added_by":"auto","created_at":"2024-01-18 19:57:29","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":252514,"visible":true,"origin":"","legend":"\u003cp\u003eRecruitment pathway for patients in this study.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/e251ebc4b497ff184fdef97f.png"},{"id":49838710,"identity":"348652a1-fe18-4b53-93e9-e5307c028439","added_by":"auto","created_at":"2024-01-18 20:05:28","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":465539,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea\u003c/strong\u003e Correlogram illustrating the auto-and cross-correlation of the 27 most-important features for classifying metastatic and normal bone tissue. The features were recorded after hierarchical clustering (nine rectangular boxes). White points indicate a positive correlation; black points indicate a negative correlation. The smaller the points and the lighter the color, the lower the correlation between two variables. \u003cstrong\u003eb \u003c/strong\u003eTexture feature selection using the least absolute shrinkage and selection operator (LASSO) binary Logistic Regression model. The radiomics signatures’ performances were explored on the receiver operating characteristics (ROC) curve. Tuning parameter (λ) selection in the LASSO model used fivefold cross-validation via a minimum. \u003cstrong\u003ec \u003c/strong\u003eLASSO coefficient profiles of the 100 bone-metastases-related texture features. A coefficient profile plot was produced against the log (λ) sequence. A vertical line is drawn at the value chosen by fivefold cross-validation.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/3993703b9ef9f573f0485d29.png"},{"id":49838709,"identity":"3dde535f-481f-4058-856a-65ac0bb6019b","added_by":"auto","created_at":"2024-01-18 20:05:28","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":93679,"visible":true,"origin":"","legend":"\u003cp\u003eReceiver operating characteristic (ROC) curves of the five models in the training and validation cohorts. The area under the curve (AUC) of the ROC curve is comprehensive.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/b2e1e8a17205fa677606ed49.png"},{"id":49838325,"identity":"1868a4a7-af1f-4009-8a9b-cadfcf054b9d","added_by":"auto","created_at":"2024-01-18 19:57:28","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":145238,"visible":true,"origin":"","legend":"\u003cp\u003eDecision curve analyses (DCA) of the various models.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/dc49e8dfd28f46d573914b19.png"},{"id":49838324,"identity":"4dfec665-1162-439d-b956-28916226af2e","added_by":"auto","created_at":"2024-01-18 19:57:28","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":136800,"visible":true,"origin":"","legend":"\u003cp\u003eCalibration curves of the various models. Four models’ calibration curves approximated ideal curves, indicating acceptable goodness-of-fit and predictive ability.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/5905c77d7c695a00d13e1837.png"},{"id":49838817,"identity":"173a7032-4bca-42f5-85db-fe50c0221dfc","added_by":"auto","created_at":"2024-01-18 20:13:29","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1295456,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3869436/v1/2628d44d-155c-4330-b3ef-83793871c67b.pdf"}],"financialInterests":"","formattedTitle":"Establishment and validation of novel predictive models to predict bone metastasis in newly diagnosed prostate adenocarcinoma based on single-photon emission computed tomography radiomics.","fulltext":[{"header":"Introduction","content":"\u003cp\u003eIn 2020, prostate cancer (PCa) was the second most common male malignancy around the world, with potentially devastating effects on the male genitourinary system\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. -oOver 90% of patients with early-stage cancer survive 5 years after diagnosis; nevertheless, once metastasis occurs, the chances of survival drop dramatically to \u0026lt;\u0026thinsp;40% within 5 years\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e, as bone lesions are rarely eradicated completely. Notably, PCa is the only solid tumor where bone metastasis (BM) precede visceral metastases. Patients with BMs survival rates are \u0026lt;\u0026thinsp;10% after 5 years. Common sites for bone metastases should be monitored in patients with PCa to maximize the effects of treatment\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eBecause of its invasive nature and high procedural risks (such as damage to the vertebral artery or spinal cord), bone biopsies are rarely used for clinical diagnosis and treatment; however, they are the gold-standard for identifying benign and malignant lesions. There is an urgent and unmet need for a non-invasive method of distinguishing benign from malignant bone lesions. In patients with bone lesions, \u003csup\u003e99m\u003c/sup\u003eTc-labeled methylene diphosphonate (\u003csup\u003e99m\u003c/sup\u003eTc-MDP) whole-body scans have been used, but their specificity was low\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. Despite undergoing imaging tests such as computed tomography (CT), magnetic resonance imaging (MRI), or bone scans, disease recurrence may be diagnosed late due to the low sensitivity of these techniques\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSingle-photon emission computed tomography/computed tomography (SPECT/CT) combines anatomical and metabolic functions to improve anatomic localization and bone imaging specificity\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. The imaging characteristics of bone metastases and benign bone lesions were similar, according to several research studies\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e. After SPECT/CT examinations, 14.3% of patients still had an equivocal diagnosis. It was difficult to distinguish between bone metastases and benign bone lesions\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. Additionally, there was no way to quantify the intensity, uniformity, or heterogeneity of lesion distribution, so SPECT/CT diagnosis depends on physicians' experience, which was always subjective\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eHealthcare is being transformed by artificial intelligence (AI), cognitive tasks performed by computers and based on available data. AI can be used to identify unique or complex imaging features and facilitates quantitative evaluation\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. In machine learning (ML), a subfield of AI, an algorithm is used to mathematically analyze the distribution of pixels (voxels) in medical gray-scale images, which can extract a large number of features from medical images in high throughput. Automatic or semi-automatic analysis methods are used to convert imaging data into a mining data space to quantify the heterogeneity of lesions. It has good application in differential diagnosis, histological classification, treatment options and prognosis analysis of diseases. There is growing interest in the use of AI to aid physicians who assess lesions. AI can surpass subjective visual interpretation in order to obtain additional information about tumor behavior and pathophysiology that is otherwise not inferable by the human eye and commonly used technologies. Medical professionals who visually evaluate a large number of uncertain images will unavoidably make mistakes. In order to improve the survival rate of patients with PCa, accurate predictive models are needed.\u003c/p\u003e \u003cp\u003eRadiomics models derived from SPECT/CT images can effectively discriminate between vertebral bone metastases and benign bone disease\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. Unfortunately, only a few studies have used SPECT/CT radiomics to study bone metastases associated with PCa. We investigated the feasibility of SPECT/CT imaging-based radiomics for identifying and diagnosing bone metastases from benign lesions in patients with PCa, mining more accurate digital information from the target area of SPECT images, and seeking more accurate guidance for patient treatment.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Population selection\u003c/h2\u003e \u003cp\u003eWe retrospectively analyzed a dataset of SPECT/CT images from 176 patients with biopsy-proven prostatic carcinoma between June 2016 and June 2022. This retrospective study was approved by the institutional review board, and the requirement for informed consent was waived. The patients were grouped randomly into the training (n\u0026thinsp;=\u0026thinsp;140) and validation (n\u0026thinsp;=\u0026thinsp;36) cohorts, with a ratio of 8:2. The major criteria for inclusion in the study were as follows: a)Eligible patients with histologically on-set confirmed PCa) All patients received a pre-treatment workup with imaging studies; c)clinical and imaging data were complete and can be followed up by telephone. We excluded: 1) patients with a history of prostate surgery, chemoradiotherapy, hormone therapy, and targeted therapy; 2) Patients with a history of malignant tumors; 3)images were unable to be evaluated owing to unsatisfactory quality; 4) region of interest (ROI) cannot be outlined clearly and correctly. Details concerning the flow of studies through the selection process are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Baseline clinicopathologic data, including age, prostate-specific antigen (PSA) level, Gleason score (GS), and histological type, were derived from electronic medical records. Bone status, including the presence or absence of BMs, was clearly determined by later follow-up outcome and MR imaging data. ECT for bone metastases was characterized by multiple radionuclide concentration areas\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e, we select 1\u0026ndash;3 typical lesions (mainly in axial bones, limb bones, pelvis, etc.) from multiple lesions to extract features [16]. Benign lesions are also selected radionuclide concentration areas, mainly in the limb bones, ribs, pelvis, according to imaging follow-up and clinical progress.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eLaboratory analysis of PSA was performed via routine blood tests within 1 week before treatment initiation or ultrasound-guided biopsy. Total serum PSA level thresholds were \u0026le;\u0026thinsp;4 ng/mL, 4\u0026thinsp;~\u0026thinsp;100 ng/mL, and \u0026gt;\u0026thinsp;100 ng/mL, which was determined using radioimmunoassay methods for determining total PSA by Quest Diagnostics and in accordance with the normal range used at our institution.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Image acquisition and segmentation\u003c/h2\u003e \u003cp\u003eAll the acquisition procedures were completed on a 16\u0026shy;section SPECT/CT scanner (dual head gamma camera Symbia Intevo bold, Siemens, Erlangen, Germany). The WBS was acquired within 2h after intravenous administration of 15\u0026ndash;25 mCi \u003csup\u003e99m\u003c/sup\u003eTc-MDP (Shanghai Syncor Medicine Corp. Ltd. (Soochow, China)), then SPECT/CT was performed immediately for further diagnosis if a suspicious lesion was found on the WBS. With a 256\u0026times;256 matrix by using a magnification of 1.00\u0026times;, the SPECT images were acquired with a 15% window centered around a 140-keV photopeak. CT scanning parameters were as follows: tube voltage, 130 kV; automatic mAs control (reference mAs 120); rotation time, 0.6 seconds; pitch, 0.8; detector configuration, 1.0 mm; and beam width, 10 mm. Coronal and sagittal views were reconstructed using 2 mm thick slices. The CT reconstruction used a hard filter (B80s). Image reconstruction program was carried out in the SYngo workstation (Syncor Medicine Corp. Ltd., Soochow, China). The image fusion program was Symbia.net.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Image analysis and expert human qualitative classification\u003c/h2\u003e \u003cp\u003eAfter summarizing all available clinical data to determine the nature of the lesion, we concluded that the study\u0026rsquo;s diagnostic criteria were based on the comprehensive pathological biopsy, laboratory indicators, follow-up imaging, and clinical progression. All SPECT images were independently evaluated by two experienced readers (MBM and XYC) with more than 10 years of experience in qualitative classification and diagnosis. The lesions were visually classified in a double-blind manner, without access to pathological test results. The readers were only given the Total Prostate-Specific Antigen (t-PSA) and Gleason scores of the patient being read. The human expert diagnostic findings were evaluated using weighted Kappa statistics to obtain interobserver agreement. The main criteria for the human expert's qualitative classification of bone metastases were bone changes on SPECT images and abnormal uptake of \u003csup\u003e99 m\u003c/sup\u003eTc-MDP in the corresponding areas.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4.Lesion segmentation and feature extraction\u003c/h2\u003e \u003cp\u003eThe patients\u0026rsquo; SPECT images were imported into an open source three-dimensional (3D) slicer V4.13.0 (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.slicer.org/\u003c/span\u003e\u003cspan address=\"https://www.slicer.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) software in DICOM format. Without knowing the pathological results, the region of abnormal 99mTc-MDP uptake on each image was described as a region of interest (ROI) by two radiologists with more than 10 years of work experience. The radiologists carefully identified the edge of the lesion and delineated along the edge of the lesion in SPECT images. And they are distributed in different parts. To assess interobserver reliability for ROI delineation, 30 lesions were randomly selected for secondary outlining\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. Before extracting the target features, the images were normalized using \u0026micro;\u0026thinsp;\u0026plusmn;\u0026thinsp;3σ (where \u0026micro; is the average value of the image gray value and σ is the SD of the image gray value) to reduce the impact of brightness and contrast on the images\u0026rsquo; gray value.\u003c/p\u003e \u003cp\u003eAfter segmenting the images, we used PyRadiomics [version 3.0.1, van Griethuysen, J. J. M., Fedorov, A., Parmar, C., Hosny, A., Aucoin, N., Narayan, V., Beets-Tan, R. G. H., Fillon-Robin, J. C., Pieper, S., Aerts, H. J. W. L. (2017). Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Research, 77(21), e104\u0026ndash;e107] to analyze the ROI. We divided candidate features and the radiomics parameters into the following categories: First Order Statistics, Shape-based, Gray Level Cooccurence Matrix, Gray Level Run Length Matrix, Gray Level Size Zone Matrix, Neigbouring, Gray Tone Difference Matrix, Gray Level Dependence Matrix and wavelet. Detailed information about radiomics features was presented previously\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e2.5. Radiomics feature Selection and ML Model Establishment\u003c/h2\u003e \u003cp\u003eWe used the Minimum-Redundancy Maximum-Relevance (mRMR)algorithm to determine which radiomic features were associated with differential diagnosis between benign and malignant lesion\u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e. The mRMR algorithm identifies features that are highly correlated with bone status, but minimally correlated with other features in order to reduce overfitting of the model. The LASSO method is suitable for the regression of high-dimensional data and can be used to select meaningful features using the non-zero coefficient. This process included a 5\u0026ndash;fold cross-validation scheme to minimize selection bias. When constructing the model, features\u0026rsquo; importance was determined by their selection probabilities. Image features with a selected frequency\u0026thinsp;\u0026gt;\u0026thinsp;90% were used to re-fit the final imaging model. Finally, four supervised ML algorithms were implemented in this study: LogisticRegression, RandomForest, Xgboost, and CatBoost. In order to obtain high accurate models, we assessed model performance using receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC) in the training and validation cohorts, respectively. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e summarizes the models\u0026rsquo; ROC curves.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Statistical analysis\u003c/h2\u003e \u003cp\u003eIBM SPSS (version 25.0) software was used for the statistical analyses, and \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant. For determining between-group differences for categorical variables, we used the chi-square test for two independent samples; the Mann-Whitney U test for two independent datasets was used for continuous variables. Weighted Kappa statistics were used to evaluate interobserver agreement. We used AUC and ROC curves to evaluate the predictive abilities of radiomic and image features. The DeLong test was used to detect statistical differences among the five predictive models.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Patients\u0026rsquo; clinicopathologic features\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e2\u003c/span\u003e is a detailed study flow chart. We enrolled 176 patients with PCa, including 81 patients with (mean age 75.27\u0026thinsp;\u0026plusmn;\u0026thinsp;7.85 years) and 95 patients without BM (mean age 73.20\u0026thinsp;\u0026plusmn;\u0026thinsp;9.11 years). Of these, 140 patients were assigned to the training set, and 36 to the validation set. There were no statistically significant differences in age (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.11) or t-PSA distribution (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.137) between the training (n\u0026thinsp;=\u0026thinsp;140) and validation (n\u0026thinsp;=\u0026thinsp;36) cohorts. Gleason grade and histological type were significantly related to BMs in both groups (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The patients\u0026rsquo; clinical characteristics are included in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eRisk factors for bone metastasis in the training and validation cohorts\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBM(+)1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNone-BM(-)2\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eP\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAge,mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD,years\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e75.27\u0026thinsp;\u0026plusmn;\u0026thinsp;7.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e73.20\u0026thinsp;\u0026plusmn;\u0026thinsp;9.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.11\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGleason score(biopsy)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGS(\u0026le;\u0026thinsp;7)1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7(8.53)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32(34.41)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGS(4\u0026thinsp;+\u0026thinsp;4\u0026thinsp;=\u0026thinsp;8)2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e35(42.68)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e30(32.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGS(9\u0026ndash;10)3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40(48.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e30(32.56)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1(1.08)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003et-PSA level(ng/ml)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026le;4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3(3.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4(4.21)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4-100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29(35.80)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e65(68.42)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e49(60.49)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e15(15.79)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11(11.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHistological type\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u0026lt;\u0026thinsp;0.001\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdenocarcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e28(34.57)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18(18.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcinar adenocarcinoma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e52(64.20)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e75(78.95)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1(1.23)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2(2.11)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eNOTE.P value is derived from the univariable association analyses between each of the clinicopahologic variables and Bone status. *P value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered to be statistically significant.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Construction of radiomics tags\u003c/h2\u003e \u003cp\u003eAfter removal by high correlation, we extracted 218 features from each ROI. We then constructed four different models based on corresponding features using LASSO; the models were compared using the AUC values, as shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003eB and C. According to the weighting coefficient corresponding to the feature (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e3\u003c/span\u003eA), a radiomics formula was obtained and used to calculate the histological score for each lesion in the training and verification groups. The radiomic features and corresponding coefficients are included in Supplementary.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.3. Performance of different prediction models\u003c/h2\u003e \u003cp\u003eThe four radiomic models in the training cohort were constructed based on different factors, with AUC values ranging from 0.87 to 0.98. The weighted Kappa value was 0.611, indicating good interobserver agreement. The AUC values of qualitative classification by human experts were 0.655 and 0.872 in the training and validation cohorts, respectively. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e includes information on the models\u0026rsquo; accuracy and the ROC curves are included in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e4\u003c/span\u003e. Logistic Regression showed the highest AUC values in the training and test groups (0.984, 0.916).The DeLong test was used for pairwise comparisons of the five prediction models to determine if any between-model differences in AUC were statistically significant. Significant differences were observed in the AUC between human experts' qualitative classification and the others (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05); however, there were no significant differences among the Logistic Regression, CatBoost, and Xgboost models (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). We used the DCA curve to verify the models\u0026rsquo; net benefit for predicting BMs. When the decision curve threshold was set at 0\u0026ndash;1, the predictive models always outperformed the human experts. (Of note, the Logistic Regression model had a slightly higher clinical gain than the other models.) The decision curve was shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003e The diagnostic ability of each model for discriminating Bone metastasis of prostate cancer from benign bone lesions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"14\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c12\" colnum=\"12\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c13\" colnum=\"13\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c14\" colnum=\"14\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c7\" namest=\"c2\"\u003e \u003cp\u003eThe training cohort\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"6\" nameend=\"c14\" namest=\"c9\"\u003e \u003cp\u003eThe validation cohort\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c12\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c13\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c14\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLogistic Regression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c13\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c14\"\u003e \u003cp\u003e0.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRandom Forest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c13\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c14\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eXgboost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c13\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c14\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCatBoost\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.98\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.99\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c13\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c14\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eHuman experts\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.67\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.66\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e0.65\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c11\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c12\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c13\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c14\"\u003e \u003cp\u003e0.71\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"14\" nameend=\"c14\" namest=\"c1\"\u003e \u003cp\u003eAUC,the area under the ROC curve; PPV ,postive predictive value; NPV ,negative predictive value.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe models\u0026rsquo; curves were nearly ideal, indicating that the models had superior adaptability and predictive ability. The calibration curve was shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. Collectively, the predictive models were accurate and outperformed human experts. Importantly, the Logistic Regression model provided the optimal prediction.\u003c/p\u003e "},{"header":"Discussion","content":"\u003cp\u003eGiven the emerging focus on patient-centered therapies, radiomics can help oncologists bridge the gap between histological findings and real microenvironment heterogeneity. Radiomics may be able to mitigate the limitations of subjective imaging-based evaluations and provide objective assessments of cancer heterogeneity while quantifying patients\u0026rsquo; survival. We built and validated a radiomics and human expert model able to distinguish BMs from normal bone in patients with PCa based on SPECT/CT images. The ML model achieved excellent binary classification performance in patients with PCa. This was one of the first SPECT imaging studies of PCa bone metastases. We also created five prediction models, rather than the 1\u0026ndash;3 models produced by previous studies. We selected a whole-body scan strategy to extract features that minimized complexity while achieving effective radiomics. With the help of image characterization algorithms, radiomic features can be quantified as criteria for determining tumor heterogeneity\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eOur proposed models provide are easy-to-use, quantitative, and individualized tools for predicting BMs. Use of our models may help avoid inappropriate surgical procedures for occult BM-positive patients while reducing the risk of BMs for high-risk patients. We used LASSO to reduce the regression coefficient and construct the radiomic signature. Our final models included the 27 best potential predictors of BMs.\u003c/p\u003e \u003cp\u003eImportantly, the radiomics signature was a better predictor of BMs than independent clinical risk factors. This suggests that, compared with gold-standard methods, the radiomics method was better at predicting BM risk. A prior multi-factor regression analysis revealed that radiomic features and GS independently predicted bone volume in patients with PCa. The GS is used to evaluate tumor aggressiveness\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe GS is used to assess risk in patients with PCa. The score could also be used to guide ML interpretation of SPECT images obtained from patients with BMs. Meanwhile, t-PSA may be a more effective predictor of PCa. However, the correlation between t-PSA and BM was not significant in this study. Several factors may affect the accuracy of biopsy t-PSA levels, including inappropriate handling prior to blood collection, hormone therapy, and inflammation.\u003c/p\u003e \u003cp\u003eRadiomics seeks to develop clinical models to identify and assess patients' condition and streamline clinical management. Most prior studies used CT/MRI imaging features to establish BM prediction models. Of the studies that used radiomics for predicting bone disease, differentiating benign from malignant bone lesions, and characterizing malignant bone lesions, 85% were MRI-based and 15% were CT-based\u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. Wang et al. used multiparametric PCa MRI to extract radiomic features, which were combined with free PSA levels and Gleason scores to predict the presence/absence of BMs in patients with PCa\u003csup\u003e[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e. The AUC of Wang\u0026rsquo;s model was 0.93 compared to 0.98 for our best model. Hinzpeter et al. proposed classifying patients' bones as either BM or normal bone using a training dataset. The researchers created a gradient-enhanced tree from 11 selected radiological features measured by quantitative methods from CT images\u003csup\u003e[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e. The training model had a classification accuracy of 0.85 (95% confidence interval: 0.76\u0026mdash;0.92, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.001), a sensitivity of 78%, and a specificity of 93%. Several of our models\u0026rsquo; strengths warrant mentioning. First, we used common SPECT features to assess BM in patients with PCa and provided clinicians with a quantitative, easy-to-use tool to predict BM risk. This effort culminated in the creation of four radiomic models. Despite this, the evidence supporting the prediction of BMs in patients with PCa using SPECT radiomic features was insufficient. A recent study by Jin et al. \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003eevaluated a radiomics model based on CT and SPECT images to discriminate between vertebral BMs and benign bone disease. BMs that occur secondary to PCa cancer are well-known. Researchers continue to mine SPECT databases for features that distinguish benign and malignant bone tumors. However, the associated cases based on characterization and ML generally suffer from low accuracy. Our study model AUC range was 0.88\u0026ndash;0.93, considered acceptable given the inherent characteristics of SPECT image textures (low resolution and image clarity). All ML models outperformed human experts for distinguishing BMs from benign bone in the training and validation sets.\u003c/p\u003e \u003cp\u003eCompared to gold-standard methods of image assessment, radiomics is a novel tool for extracting image information using high-throughput methods\u003csup\u003e[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e.The resultant information is used for disease typing and grading, gene localization, early treatment, and prognostication. Some studies have shown that radiomics performs better than traditional clinical methods for non-invasive classification and diagnosis of diseases\u003csup\u003e[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e. Expanded use of SPECT may allow most patients with PCa to detect bone metastases early and evaluate treatment efficacy given that SPECT is more sensitive than MRI or PET/CT and widely available to patients.\u003c/p\u003e \u003cp\u003eThere is an obvious need for continued research in this area. Because our findings are only applicable to PCa, our results cannot be generalized to patients with other diseases. Future, well-powered studies are needed to validate use of our models with patients carrying other diagnoses.\u003c/p\u003e \u003cp\u003eAll lesions\u0026rsquo; statuses were confirmed by biopsy or follow-up data; if the final diagnosis was uncertain, the lesion was simply removed. Although time-consuming, these exclusions ensure our study\u0026rsquo;s rigor. Finally, we excluded patients who received treatment because a certain percentage of patients experience flare-ups and osteogenic reactions after chemotherapy or radiotherapy, which may also affect the radiotracer uptake.\u003c/p\u003e \u003cp\u003eOur study has several limitations that warrant consideration. This was a retrospective, and not prospective, research design. Our study cohort was relatively small; this was the most critical limitation of this study. We did not validate our findings using external data and there is most certainly a selection bias. Non-uptake areas on SPECT were considered normal bone. Lastly, the radiomic features we identified may not be generalizable to patients with other diseases. In the end, we were unable to perform a detailed histopathological analysis of each case; BMs and benign bone lesions were primarily confirmed based on pathological findings, radiographic follow-up, and disease progression. PCa radiomics should emerge as a consistent and objective tool for setting up clinical trials and tailoring treatment, as it allows for accurate assessment of patients at the time of diagnosis and during treatment.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eModels developed using different algorithms and radiomics features performed better than human experts at distinguishing benign bone from BMs secondary to PCa. These findings suggest that AI models can be applied to clinical settings. The Logistic Regression model was best predicting BMs.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the guidelines of the Declaration of Helsinki. The studies involving human participants were reviewed and approved by Ethics Committee of The Second Affiliated Hospital of Soochow University. The ethics committee waived the requirement of written informed consent for participation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConception and design: NW and SQ. Collection and assembly of the data: WK. Development of the methodology: QH. Data analysis and interpretation: ZH. Manuscript writing: All authors. Manuscript review: YS and ZL. All authors contributed to the article and approved the submitted version.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 2021;71(3):209\u0026ndash;49.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCulp MB, Soerjomataram I, Efstathiou JA, Bray F, Jemal A. Recent Global Patterns in Prostate Cancer Incidence and Mortality Rates. Eur Urol. 2020;77(1):38\u0026ndash;52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNorum J, Nieder C. Treatments for Metastatic Prostate Cancer (mPC): A Review of Costing Evidence. Pharmacoeconomics 2017, \u003cem\u003e35\u003c/em\u003e (12), 1223\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChaffer CL, Weinberg RA, Perspective A. on Cancer Cell Metastasis. Science 2011, \u003cem\u003e331\u003c/em\u003e (6024), 1559\u0026ndash;1564.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSvensson E, Christiansen CF, Ulrichsen SP, Rorth MR, Sorensen HT. Survival after bone metastasis by primary cancer type: a Danish population-based cohort study. BMJ Open 2017, 7 (9), e016022.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, Sargent D, Ford R, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45(2):228\u0026ndash;47.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Z, Chen X, Wang R. Application of SPECT and PET / CT with computer-aided diagnosis in bone metastasis of prostate cancer: a review. Cancer Imaging. 2022;22(1):18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCalais J, Cao M, Nickols NG. The Utility of PET/CT in the Planning of External Radiation Therapy for Prostate Cancer. J Nucl Med. 2018;59(4):557\u0026ndash;67.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlqahtani MM, Fulton R, Constable C, Willowson KP, Kench PL. Diagnostic performance of whole-body SPECT/CT in bone metastasis detection using (99m)Tc-labelled diphosphate: a systematic review and meta-analysis. Clin Radiol 2020, \u003cem\u003e75\u003c/em\u003e (12), 961 e11-961 e24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLangsteger W, Rezaee A, Pirich C, Beheshti M. (18)F-NaF-PET/CT and (99m)Tc-MDP Bone Scintigraphy in the Detection of Bone Metastases in Prostate Cancer. Semin Nucl Med. 2016;46(6):491\u0026ndash;501.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAbikhzer G, Srour S, Keidar Z, Bar-Shalom R, Kagna O, Israel O, et al. Added Value of SPECT/CT in the Evaluation of Benign Bone Diseases of the Appendicular Skeleton. Clin Nucl Med. 2016;41(4):e195\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuezennec C, Keromnes N, Robin P, Abgral R, Bourhis D, Querellou S et al. Incremental diagnostic utility of systematic double-bed SPECT/CT for bone scintigraphy in initial staging of cancer patients. Cancer Imaging 2017, \u003cem\u003e17\u003c/em\u003e (1), 16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDe Schepper S, Gnanasegaran G, Dickson JC, Van den Wyngaert T. Absolute Quantification in Diagnostic SPECT/CT: The Phantom Premise. Diagnostics 2021, \u003cem\u003e11\u003c/em\u003e (12).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosny A, Parmar C, Quackenbush J, Schwartz LH, Aerts H. Artificial intelligence in radiology. Nat Rev Cancer 2018, \u003cem\u003e18\u003c/em\u003e (8), 500\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJin Z, Zhang F, Wang Y, Tian A, Zhang J, Chen M, et al. Single-Photon Emission Computed Tomography/Computed Tomography Image-Based Radiomics for Discriminating Vertebral Bone Metastases From Benign Bone Lesions in Patients With Tumors. Front Med (Lausanne). 2021;8:792581.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXue C, Yuan J, Lo GG, Chang ATY, Poon DMC, Wong OL, et al. Radiomics feature reliability assessed by intraclass correlation coefficient: a systematic review. Quant Imaging Med Surg. 2021;11(10):4431\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZwanenburg A, Vallieres M, Abdalah MA, Aerts H, Andrearczyk V, Apte A, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020;295(2):328\u0026ndash;38.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePeng HC, Long FH, Ding C. Feature selection based on mutual information: Criteria of max-dependency, max-relevance, and min-redundancy. Ieee Trans Pattern Anal Mach Intell 2005, \u003cem\u003e27\u003c/em\u003e (8), 1226\u0026ndash;38.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLimkin EJ, Sun R, Dercle L, Zacharaki EI, Robert C, Reuze S, et al. Promises and challenges for the implementation of computational medical imaging (radiomics) in oncology. Ann Oncol. 2017;28(6):1191\u0026ndash;206.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrunese L, Mercaldo F, Reginelli A, Santone A. Radiomics for Gleason Score Detection through Deep Learning. Sens (Basel). 2020;20:18.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFaiella E, Santucci D, Calabrese A, Russo F, Vadala G, Zobel BB et al. Artificial Intelligence in Bone Metastases: An MRI and CT Imaging Review. Int J Environ Res Public Health 2022, \u003cem\u003e19\u003c/em\u003e (3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Y, Yu B, Zhong F, Guo Q, Li K, Hou Y, et al. MRI-based texture analysis of the primary tumor for pre-treatment prediction of bone metastases in prostate cancer. Magn Reson Imaging. 2019;60:76\u0026ndash;84.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHinzpeter R, Baumann L, Guggenberger R, Huellner M, Alkadhi H, Baessler B. Radiomics for detecting prostate cancer bone metastases invisible in CT: a proof-of-concept study. Eur Radiol. 2022;32(3):1823\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTomaszewski MR, Gillies RJ. The Biological Meaning of Radiomic Features. Radiol 2021, \u003cem\u003e298\u003c/em\u003e (3), 505\u0026ndash;16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLohmann P, Franceschi E, Vollmuth P, Dhermain F, Weller M, Preusser M, et al. Radiomics in neuro-oncological clinical trials. Lancet Digit Health. 2022;4(11):e841\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang M, Hu P, Li M, Ding R, Wang Y, Pan S, et al. Computed Tomography-Based Radiomics in Predicting T Stage and Length of Esophageal Squamous Cell Carcinoma. Front Oncol. 2021;11:722961.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"annals-of-nuclear-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"anme","sideBox":"Learn more about [Annals of Nuclear Medicine](http://link.springer.com/journal/12149)","snPcode":"12149","submissionUrl":"https://www.editorialmanager.com/anme/default2.aspx","title":"Annals of Nuclear Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"prostate adenocarcinoma, radiomics, single-photon emission computed tomography, bone metastasis","lastPublishedDoi":"10.21203/rs.3.rs-3869436/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3869436/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003ePurpose\u003c/h2\u003e \u003cp\u003eIn order to establish and validate novel predictive models for predicting bone metastasis (BM) in newly diagnosed prostate adenocarcinoma (PCa) in single-photon emission computed tomography radiomics.\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eIn a retrospective review of clinical SPECT database, 176 patients (training set: n\u0026thinsp;=\u0026thinsp;140; validation set: n\u0026thinsp;=\u0026thinsp;36) who underwent SPECT/CT imaging and histologically confirmed with newly diagnosed PCa from June 2016 to June 2022 were enrolled. Radiomic features were extracted from ROI in a targeted lesion of each patient. Clinical features, including age, t-PSA, and Gleason grades, were included. Statistical tests were then used to eliminate irrelevant and redundant features. Finally, three types of optimized models were constructed for the prediction. Furthermore, 5-fold cross-validation was applied to obtain the sensitivity, specificity, accuracy, and area under the curve (AUC) for performance evaluation. The clinical usefulness of the multivariate models was estimated through decision curve analysis (DCA).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eRadiomics signature consisting of 27 selected features was significantly correlated with bone status(P\u0026thinsp;\u0026lt;\u0026thinsp;0.01 for both training and validation sets). Collectively, the models showed good predictive efficiency. The AUC values ranged from 0.87 to 0.98 in four models. The AUC value of the human experts was 0.655 and 0.872 in the training and validation groups, respectively. Most radiomic models showed better diagnostic accuracy than human experts in the training group and the validation group. DCA also demonstrated the superiority of the radiomics models compared to human experts.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eOur proposed models, which incorporate SPECT-based radiomics signature and clinical risk factors, could be a promising auxiliary means to assist radiologists or medical physicians in their subsequent workup to confirm the diagnosis of BM.\u003c/p\u003e","manuscriptTitle":"Establishment and validation of novel predictive models to predict bone metastasis in newly diagnosed prostate adenocarcinoma based on single-photon emission computed tomography radiomics.","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-18 19:57:24","doi":"10.21203/rs.3.rs-3869436/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"reviewerAgreed","content":"","date":"2024-01-17T11:01:56+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-01-16T11:52:13+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-01-16T04:12:59+00:00","index":"","fulltext":""},{"type":"submitted","content":"Annals of Nuclear Medicine","date":"2024-01-15T20:09:16+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"annals-of-nuclear-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"anme","sideBox":"Learn more about [Annals of Nuclear Medicine](http://link.springer.com/journal/12149)","snPcode":"12149","submissionUrl":"https://www.editorialmanager.com/anme/default2.aspx","title":"Annals of Nuclear Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"cdfaea5c-58f6-4a61-bbd3-c20591a5354c","owner":[],"postedDate":"January 18th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2024-05-12T22:57:11+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-18 19:57:24","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3869436","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3869436","identity":"rs-3869436","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}