Distinguishing Lymphoma from Benign Lymph Node Diseases in Fever of Unknown Origin using PET/CT Radiomics

doi:10.21203/rs.3.rs-4625440/v1

Download PDF

Research Article

Distinguishing Lymphoma from Benign Lymph Node Diseases in Fever of Unknown Origin using PET/CT Radiomics

https://doi.org/10.21203/rs.3.rs-4625440/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background A considerable portion of patients with fever of unknown origin (FUO) present concomitant lymphadenopathy. Diseases within the spectrum of FUO accompanied by lymphadenopathy include lymphoma, infections, and rheumatic diseases. Particularly, lymphoma has emerged as the most prevalent etiology of FUO with associated lymphadenopathy. However, conventional imaging techniques, including PET/CT, often struggle to accurately distinguish between malignant and benign lymph node lesions. This study utilizes PET/CT radiomics to differentiate between malignant and benign lymph node lesions in patients with FUO, aiming to improve the accuracy of diagnosing lymphoma and benign lymph node diseases.

Results Data were collected from 204 patients who underwent ¹⁸F-FDG PET/CT examinations for FUO, including 114 lymphoma patients and 90 patients with benign lymph node lesions. Patients were randomly divided into training and testing groups at a ratio of 7:3. A total of 15 effective features were obtained by the least absolute shrinkage and selection operator (LASSO) algorithm. Machine learning models were constructed using logistic regression (LR), support vector machine (SVM), random forest (RF), and k-nearest neighbors (KNN) algorithms. In the training group, the AUC values for predicting benign and malignant cases by LR, SVM, RF, and KNN models were 0.936, 0.930, 0.998, and 0.938, respectively. There were statistically significant differences in AUC between the RF and other models (all P < 0.001). In the testing group, the AUC values for the four models were 0.860, 0.866, 0.915, and 0.891, respectively, with no statistically significant differences found between the four models (all P > 0.05). The DCA curves of the RF model outperformed those of the other three models in both the training and testing groups.

Conclusions PET/CT radiomics demonstrates promising performance in discriminating lymphoma from benign lymph node lesions in patients with FUO, with the RF model showing the best performance in distinguishing between lymphoma and benign lymph node diseases.

Fever of unknown origin

Lymphoma

Lymph Node Diseases

18F-FDG PET/CT

Radiomics

Healthcare practitioners have developed a comprehensive understanding of fever of unknown origin (FUO) over an extended period of time. In 1961, Petersdorf and Beeson defined FUO as "a condition characterized by a fever of 38.3°C or higher, measured sublingually, persisting for more than 3 weeks, with the cause remaining unidentified even after hospitalization for 1 week or longer" [1, 2]. In 1991, Durack and Street revised the criteria for FUO as "a condition characterized by a persistent fever of 38.3°C or higher of unknown origin lasting for more than 3 weeks, which remains undiagnosed despite 3 days of inpatient examination or 3 outpatient examinations" [3]. FUO is divided into four categories: classic (the most common category), nosocomial, neutropenic, and HIV-associated. The etiology of classic FUO is classified into five categories: infection, neoplasms, inflammation, other causes, and undiagnosed reasons [4, 5]. Diverse living environments, geographical locations, and care statuses contribute to variations in the spectrum of diseases; however, infections and neoplasms remain the most common etiologies for FUO overall. Approximately 12%-35% of patients ultimately succumb to infections or neoplasms related to FUO. Prognosis for those FUO patients whose etiology remains undetermined tends to be relatively favorable [6, 7].

A considerable portion of patients with FUO present concomitant lymphadenopathy. Diseases within the spectrum of FUO accompanied by lymphadenopathy include lymphoma, infections, and rheumatic diseases, all common causes of FUO. Particularly, lymphoma has emerged as the most prevalent etiology of FUO with associated lymphadenopathy [8, 9]. The timely diagnosis of lymphoma in FUO patients is crucial for treatment and prognosis. However, conventional imaging assessments of lymph node size and morphology are not reliable diagnostic standards, as the diagnosis of lymphoma primarily depends on histopathology. Nonetheless, histopathological evaluation often necessitates invasive procedures, which patients may sometimes be unable to tolerate. Furthermore, even upon successful completion of surgery, false-negative results are not uncommon in such procedures [10, 11]. Currently, ¹⁸F-fuorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) is considered the most accurate imaging modality for diagnosing FUO [12–15], as it allows for diagnosis from both metabolic and morphological perspectives, particularly facilitating the diagnosis, staging, and assessment of therapeutic efficacy in lymphoma. Multiple meta-analysis studies have demonstrated that the sensitivity of PET/CT in diagnosing FUO ranges from 86–98%, with specificity ranging from 52–85% [16, 17]. As previously mentioned, many FUO patients present with concurrent lymph node lesions. PET/CT exhibits high sensitivity in diagnosing lymph node lesions, but its specificity is relatively lower due to both lymphoma and benign lymph nodes showing high FDG uptake, leading to misdiagnosis and diagnostic challenges [18, 19]. Radiomics is the utilization of automated or semi-automated software to extract and analyze a large number of quantitative imaging features from medical images, aiming to construct radiomic models for in-depth analysis of clinical information [20]. Its theoretical foundation lies in regarding medical images not only as tools for visual assessment but also as valuable sources of quantitative data. PET/CT radiomics is primarily applied in tumor discrimination, prediction of genetic mutations, assessment of therapeutic efficacy, and prognosis prediction. The tumors involved include non-small cell lung cancer, breast cancer, colorectal cancer, gastric cancer, and malignant tumors of the central nervous system. Research on the etiological diagnosis of FUO using PET/CT radiomics is scarce. However, the assessment of lymph node characteristics is crucial in diagnosing the cause of FUO. Therefore, this study aims to utilize the quantitative imaging advantages of PET/CT radiomics to distinguish between malignant and benign lymph node lesions in FUO. This approach is expected to provide a more accurate assessment of lymphoma and benign lymph node diseases.

Study cohort

This study retrospectively collected data from patients who underwent ¹⁸F-FDG PET/CT scans due to FUO from January 2018 to December 2023. The diagnosis of FUO was based on the following criteria: (1) fever exceeding 38.3°C on three or more occasions, (2) duration of illness exceeding 3 weeks, and (3) an inconclusive diagnosis despite appropriate investigation, with at least three outpatient visits or at least 3 days of hospitalization. This study was approved by the Ethics Committee of Hebei General Hospital (Approval No.: 2024-LW-081).

Inclusion criteria were as follows: (1) Confirmed diagnosis of fever of unknown origin (FUO) with all patients undergoing ¹⁸F-FDG PET/CT scans. (2) PET/CT scans revealing lymph node lesions, with lymph node lesions measuring ≥ 1 cm on CT axial scans or lymph node metabolic values higher than the mediastinal blood pool [8, 9]. (3) All cases had a confirmed diagnosis, with malignant cases diagnosed through pathological biopsy and benign cases diagnosed through pathological examination or clinical follow-up. Exclusion criteria were as follows: (1) Hospital-acquired FUO, immunodeficiency-related FUO, and travel-related FUO. (2) History of previous malignancy or detection of solid tumors other than lymphoma on PET/CT scans. (3) Tumor-related treatment prior to PET/CT examination. All patients underwent definitive diagnosis through pathological biopsy or clinical follow-up.

The enrolled patients were randomly divided into two groups at a ratio of 7:3, forming the training group and the testing group, respectively.

¹⁸ F-FDG PET/CT scans

The examinations were conducted using a GE Discovery Elite PET/CT scanner, with a radiotracer purity of ¹⁸F-FDG ≥ 90%. Prior to ¹⁸F-FDG injection, participants fasted for over 6 hours, with fasting blood glucose levels ≤ 11.1 mmol/L. Intravenous injection of ¹⁸F-FDG was administered at a dosage of 3.70–5.55 MBq/kg. Before scanning, participants orally ingested 600 ml of oral contrast agent solution containing iohexol to opacify the gastrointestinal tract. Following the injection of ¹⁸F-FDG, participants rested quietly for 60 minutes before undergoing PET/CT scanning. The scanning range extended from the base of the skull to the mid-thigh, with acquisition performed over 6–7 bed positions, each lasting 2.5 minutes. 3D-time-of-flight (TOF) acquisition was utilized, with attenuation correction using full-energy X-ray attenuation and ordered subset expectation maximization (OSEM) reconstruction. Low-dose CT scanning was performed on the same scanner, with tube voltage set at 120 kV and tube current at 100 mA. CT slice thickness was set at 3.3 mm. PET and CT images were fused for visualization.

Image segmentation and extraction of radiomic features

The PET images were imported into LIFEx software (version 7.3.0; https://www.lifexsoft.org/) for manual delineation of volumes of interest (VOIs). VOIs representing lymph node lesions were delineated manually, selecting lymph nodes with axial scan ≥ 1 cm or with metabolic values higher than mediastinal blood pool. VOI delineation was performed by two nuclear medicine physicians with over 5 years of diagnostic experience. Conventional PET parameters of the lesions were recorded, including Maximum Standardized Uptake Value (SUVmax), Mean Standardized Uptake Value (SUVmean), Total Metabolic Tumor Volume (MTV), and Total Lesion Glycolysis (TLG). Radiomic features were extracted using the open-source PyRadiomics (http://www.radiomics.io/pyradiomics.html). The workflow for radiomic analysis is shown in Fig. 1.

Feature selection and model construction

Initially, Mann-Whitney U test was conducted in the training set to prevent overfitting of the model, selecting features with significant differences (P < 0.05) between the lymphoma group and the benign group. Subsequently, the least absolute shrinkage and selection operator (LASSO) algorithm, combined with ten-fold cross-validation, was utilized to select significant radiomic features with non-zero coefficients. These selected features were then employed to develop four classical machine learning methods: Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and k-Nearest Neighbor (KNN) algorithms.

Model evaluation and statistical methods

This study employed Python 3.7.16 for statistical analysis. The ability of different models to distinguish between lymphoma and benign diseases was evaluated in the testing group. Statistical methods used to assess their performance metrics include Receiver Operating Characteristic (ROC) curves, Area Under the Curve (AUC) of the ROC curve, Sensitivity, Specificity, Positive Predictive Value (PPV), Negative Predictive Value (NPV), and Accuracy. DeLong's non-parametric test was used to compare the differences in AUC values between models. Decision Curve Analysis (DCA) was utilized to evaluate the predictive value of the prediction model in the training and testing cohorts. P < 0.05 was considered statistically significant.

Clinical characteristics of patients

A total of 325 patients with FUO accompanied by lymphadenopathy were diagnosed. Among them, 89 patients remained undiagnosed at the end of follow-up, 25 patients were found to have solid organ tumors, and 7 patients had a history of malignant tumors. Finally, 204 patients were included in the study, comprising 114 lymphoma patients and 90 benign cases (The spectrum of diseases and the number of cases are shown in Table 1). There were 100 male and 104 female patients, with a mean age of 48.5 ± 20.9 years and an age range from 3 to 85 years. Patients were randomly divided into two groups at a ratio of 7:3, with 143 cases allocated to the training group and the remaining 61 cases to the testing group.

Table 1

Classification spectrum and number of cases of diseases in FUO patients
Disease Spectrum	Overall (n)	Percentage (%)
Lymphoma (114)
Diffuse Large B-cell Lymphoma	28	13.7
T-cell lymphoma	18	8.8
Marginal zone lymphoma	14	6.9
Follicular Lymphoma	13	6.4
Hodgkin's Lymphoma	12	5.9
High-grade B-cell lymphoma	6	2.9
B-cell lymphoblastic lymphoma	6	2.9
Small lymphocytic lymphoma	5	2.6
Mantle cell lymphoma	4	2.0
Unspecified lymphoma	8	3.9
Benign diseases (90)
Adult-onset Still's disease	21	10.3
Necrotizing lymphadenitis	15	7.4
Reactive lymphadenopathy (of unknown etiology)	11	5.4
Bacterial infection (non-tuberculous)	10	4.9
Tuberculosis of lymph nodes	5	2.5
Viral infection	5	2.5
Fungal infection	3	1.5
Arthritis	3	1.5
Connective tissue disease (unclassifiable)	3	1.5
Hemophagocytic syndrome	2	1.0
Sjögren's syndrome	2	1.0
Allergic fever	2	1.0
Crohn's disease	2	1.0
Hashimoto's thyroiditis	2	1.0
Rheumatic polymyalgia	1	0.5
Retroperitoneal fibrosis	1	0.5
Giant cell arteritis	1	0.5
Systemic lupus erythematosus	1	0.5

Feature extraction and model establishment

A total of 1203 radiomic features were extracted using PyRadiomics software. After screening with the Mann-Whitney U test, 978 relevant radiomic features remained. Following data standardization and LASSO dimensionality reduction, a final set of 15 non-zero coefficient features were obtained, which were considered as important predictive factors for discriminating between lymphoma and benign lymph node diseases. The analysis results of LASSO are shown in Fig. 2, and the corresponding 15 radiomic features along with their coefficients are listed in Table 2. A model was constructed using the extracted optimal features to differentiate between lymphoma and benign lymph node diseases.

Table 2

The optimal radiomic parameters extracted by the LASSO algorithm
Feature name	Coefficient
original_shape_LeastAxisLength	0.000508291
exponential_firstorder_Minimum	0.019797504
lbp-2D_firstorder_Kurtosis	-0.004751370
squareroot_gldm_LargeDependenceEmphasis	0.000407637
squareroot_gldm_LargeDependenceLowGrayLevelEmphasis	7.314223885
squareroot_glrlm_RunLengthNonUniformityNormalized	-0.373644382
wavelet-LHL_gldm_DependenceVariance	0.008808973
wavelet-LHH_firstorder_Skewness	-0.034160110
wavelet-LHH_gldm_DependenceVariance	0.000850229
wavelet-LHH_gldm_LargeDependenceLowGrayLevelEmphasis	0.000678951
wavelet-HLL_gldm_LargeDependenceLowGrayLevelEmphasis	0.000583544
wavelet-LLL_ngtdm_Coarseness	-1.089230303

Model performance evaluation

Four classical models including LR, SVM, RF, and KNN, their ROC curves in the training and testing sets are shown in the Fig. 3. The statistical indicators for discriminating between lymphoma and benign lymph node diseases based on these four basic machine learning algorithms are presented in Table 3. In the training set, the AUC value of the RF model is 0.998 (95% CI, 0.995-1.000), which is higher than that of LR, SVM, and KNN models. The DeLong test shows that there is a statistically significant difference in AUC between the RF model and the SVM, LR and KNN models (all P < 0.001), while there is no statistically significant difference in AUC between the LR, SVM, and KNN models (P = 0.859, 0.146, 0.418, respectively). In the testing set, the AUC value of the RF model is 0.915 (95% CI, 0.831–0.978), and the DeLong test shows that there is no statistically significant difference in AUC among the four models in the testing set (all P > 0.05). In the training set, the RF model demonstrates higher performance indicators (including sensitivity, specificity, PPV, NPV, accuracy) compared to LR, SVM, and KNN models. In the testing set, both the RF and LR models show similar performance in terms of sensitivity, specificity, PPV, NPV, outperforming the SVM and KNN models. Therefore, the RF model exhibits higher predictive performance in both the training and testing sets, surpassing the SVM, LR, and KNN models.

Table 3

Predictive performance of four machine learning algorithms
Models	AUC (95%CI)	Sens (%)	Spec (%)	PPV (%)	NPV (%)	Accuracy (%)
Training cohort
LR	0.936(0.897–0.969)	86.1	85.7	88.3	83.1	85.9
SVM	0.930 (0.888–0.965)	79.7	92.1	92.6	78.4	85.2
RF	0.998(0.995-1.000)	97.5	98.4	98.7	96.9	97.9
KNN	0.938(0.902–0.970)	84.8	88.9	90.5	82.4	86.6
Testing cohort
LR	0.860 (0.741–0.955)	88.6	77.8	83.8	84.0	83.9
SVM	0.866 (0.758–0.953)	82.9	77.8	82.9	77.8	80.6
RF	0.915 (0.831–0.978)	88.6	77.8	83.8	84.0	83.9
KNN	0.891 (0.802–0.959)	88.6	66.7	77.5	81.8	79.0

The DCA of the four models is shown in Fig. 4 to compare their clinical diagnostic value. The DCA demonstrates that the RF model curve is consistently higher than the curves of the other three models, indicating a higher net benefit and clinical value in predicting lymphoma versus benign lymph node diseases using the RF model.

FUO is a common clinical syndrome characterized by complex etiology and diverse manifestations, often posing challenges to clinicians in diagnosis and treatment. Patients with FUO often present with lymphadenopathy [12, 21], as lymph nodes are part of the immune system and their enlargement may be an immune response to potential infections, inflammation, tumors, and other etiologies. This reaction may be defensive against potential pathogens or abnormal cells, but sometimes it may also be part of the disease process. In FUO patients, lymphadenopathy may serve as a suggestive symptom, aiding healthcare providers in screening and diagnosing the underlying causes. Lymph node enlargement may suggest potential etiologies such as infection, inflammation, immune-related diseases, or tumors. Therefore, the observation and assessment of lymph nodes play a crucial role in the diagnostic process of FUO patients.

This study is consistent with previous research, where lymphoma is the most common malignancy disease associated with lymphadenopathy in cases of FUO [22, 23], while infections are the most common benign conditions, followed by adult Still's disease and necrotizing lymphadenitis. Indolent lymphomas (including low-grade follicular lymphoma, marginal zone lymphoma, small lymphocytic lymphoma, etc.) and some infiltrative lymphomas (such as mantle cell lymphoma, angioimmunoblastic T-cell lymphoma, etc.) sometimes present as mildly hypermetabolic lymph nodes on PET/CT, with no significant abnormalities in lymph node morphology or structure. However, some benign FUO conditions such as adult Still's disease, necrotizing lymphadenitis, and certain infectious diseases may also manifest as hypermetabolic lymph nodes on PET/CT [24], with abnormal lymph node morphology and structure. This poses a challenge for nuclear medicine physicians in diagnosing lymph node involvement associated with FUO using PET/CT, as it is difficult to distinguish the nature of lymph node lesions by visual inspection alone (For example Fig. 5).

Both this study and literature reports have found that adult Still's disease and necrotizing lymphadenitis often present as hypermetabolic lymph nodes [25], with metabolic characteristics similar to lymphoma, making them easily misdiagnosed as lymphoma. Moreover, these two diseases are relatively common in FUO cases, further complicating the PET/CT diagnosis of lymph node involvement. Therefore, this study aims to explore new diagnostic approaches for lymph node involvement associated with FUO by leveraging the machine learning advantages of radiomics. ¹⁸F-FDG PET/CT radiomics has shown promising results in tumor diagnosis, treatment response assessment and prognostic prediction [26]. However, there are currently no research results on radiomics for distinguishing the etiology of FUO, which is also an innovative advantage of this study.

In this study, radiomic features were extracted using PyRadiomics software. After screening with the Mann-Whitney U test, a total of 978 features remained, and finally, 15 effective features were obtained after LASSO processing. Four commonly used machine learning algorithms (SVM, RF, LR, KNN) were employed to differentiate between benign and malignant lymph node lesions in FUO cases. In the training set, RF demonstrated superior performance in terms of AUC and various performance metrics, with statistically significant differences. In the testing set, the performance metrics of RF were superior to those of other algorithms. However, there was no statistically significant difference in AUC comparison, possibly due to the smaller sample size in the testing set. The study indicates that the radiomics based on RF model algorithm can effectively differentiate the nature of lymph node lesions in FUO.

Chen et al. investigated the diagnostic performance of PET/CT for FUO with lymph node lesions, finding that relying solely on PET/CT imaging, the sensitivity for diagnosing lymphoma was 81.0%, with a specificity of 47.6%, indicating relatively low specificity [9]. SUVmax demonstrated high sensitivity in discriminating between benign and malignant diseases but had low specificity and diagnostic efficiency, consistent with previous research findings [27]. Therefore, determining lymph node benignity or malignancy based solely on SUVmax levels is deemed unreliable. Chen et al. found that a scoring system based on PET/CT and clinical parameters performed well in distinguishing between lymphoma and benign diseases, serving as a reliable non-invasive tool [9]. These clinical parameters included blood cell analysis, liver function analysis, lactate dehydrogenase, C-reactive protein, erythrocyte sedimentation rate, serum ferritin, procalcitonin, antinuclear antibodies, rheumatoid factors, urine analysis, interferon release assays, blood cultures, urine cultures, chest CT scans, ultrasound, and a series of other examinations and tests. Their scoring system yielded an AUC of 0.93 (95% CI, 0.89–0.97) for diagnosing lymphoma. In this study, both in the training and testing groups, the AUCs for distinguishing between lymphoma and benign lymph node diseases were 0.998 (95% CI, 0.995-1.000) and 0.915 (95% CI, 0.831–0.978), respectively. Compared to the complex array of clinical parameters and subjective scoring criteria, radiomics operations are simpler, faster, and more objective. Our results demonstrate that the diagnostic performance of radiomics in the training group surpasses that of Chen et al.'s study, albeit slightly lower in the testing group, possibly due to the relatively smaller dataset in the testing group.

When the diagnosis of FUO becomes challenging using PET/CT imaging, a radiomics model based on RF can be employed to differentiate between lymphoma or benign lymph node diseases, swiftly and accurately distinguishing the benign and malignant causes of FUO and optimizing diagnostic strategies to avoid unnecessary invasive surgeries. The distinctive feature of this study lies in leveraging the advantages of radiomics machine learning to establish a simple discriminative model, enhancing the diagnostic efficiency of PET/CT for FUO with lymph node lesions, and eliminating subjective biases. This discriminative model performs well in distinguishing between benign and malignant lesions. When clinical features of patients are atypical and diagnostic clues are lacking, the discriminative model can assist clinicians in improving the accuracy of FUO diagnosis.

This study has several limitations. Firstly, the relatively small sample size resulted in a limited number of cases in the test group, which may have interfered with the diagnostic performance. Secondly, this study is a single-center retrospective study, and future research should involve larger-scale multicenter prospective studies. Additionally, this study lacks external validation, and future studies will incorporate diverse data sources to validate the accuracy of the model.

Lymphoma is the most common malignant tumor in FUO with lymphadenopathy, while infection, adult Still's disease, and necrotizing lymphadenitis are the most common benign diseases associated with FUO and lymphadenopathy. The ¹⁸F-FDG PET/CT radiomics model demonstrates good performance in discriminating between benign and malignant causes of FUO with lymphadenopathy, with the RF model performing the best. However, the diagnostic performance of the RF model in the test set is slightly lower than that in the training set, which needs further improvement in future large-sample and multicenter studies.

FUO	Fever of Unknown Origin
LASSO	Least Absolute Shrinkage and Selection Operator
LR	Logistic Regression
SVM	Support Vector Machine
RF	Random Forest
KNN	K-Nearest Neighbors
¹⁸F-FDG PET	¹⁸F-fluorodeoxyglucose positron emission tomography
TOF	Time-of-flight
OSEM	Ordered Subset Expectation Maximization
VOI	Volumes of Interest
SUVmax	Maximum Standardized Uptake Value
SUVmean	Mean Standardized Uptake Value
MTV	Metabolic Tumor Volume
TLG	Total Lesion Glycolysis
ROC	Receiver Operating Characteristic
AUC	Area Under theCurve
PPV	Positive Predictive Value
NPV	Negative Predictive Value
DCA	Decision Curve Analysis

Ethics approval and consent to participate

All procedures performed in the study and involving human participants were carried out in accordance with the ethical standards of the institutional and/or national research committee and with the principles of the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Hebei general Hospital (2024-LW-081). This retrospective analysis was approved by the Ethics Committee of Hebei general Hospital, and the requirement of informed consent was waived.

Consent for publication

Not applicable

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no conflict of interest.

Funding

This work was supported by National Natural Science Foundation of China [82302252], Natural Science Foundation of Hebei Province [H2022307068], Key Science and Technology Research Program of the Health Commission of Hebei Province (20241560) and Key Science and Technology Research Program of the Health Commission of Hebei Province (20200016).

Authors’ Contributions

All authors contributed to the study conception and design. Material preparation and data collection were performed by XCZ and YZB. Data analysis was performed by XCZ and FLJ. YJH, CNT, JYZ, SHL, QW, KL, LZ, JLL, JJZ were responsible for the concept and design of the study. The first draft of the manuscript was written by XCZ and all authors commented on previous versions of the manuscript. XCZ revised it and YZB supervised the study. All authors read and approved the final manuscript.

Acknowledgments

The authors thank all their coworkers involved in the study for their support and assistance.

Minamimoto R. Optimal use of the FDG-PET/CT in the diagnostic process of fever of unknown origin (FUO): a comprehensive review. Japanese J Radiol. 2022;40:1121–37. https://doi.org/10.1007/s11604-022-01306-w.
Petersdorf Rg Fau -, Beeson PB, Beeson PB. Fever of unexplained origin: report on 100 cases. Med (Baltim). 1961;40:1–30. https://doi.org/10.1097/00005792-196102000-00001.
Unger M, Karanikas G, Kerschbaumer A, Winkler S, Aletaha D. Fever of unknown origin (FUO) revised. Wiener klinische Wochenschrift. 2016;128:796–801. https://doi.org/10.1007/s00508-016-1083-9.
Wright WFAP. Fever and Fever of Unknown Origin (FUO): review, recent advances, and lingering dogma. Open Forum Infect Dis. 2020;7:ofaa132. https://doi.org/10.1093/ofid/ofaa132/5828054.
Palestro CJ, Brandon DC, Dibble EH, Keidar Z, Kwak JJ. FDG PET in Evaluation of Patients With Fever of Unknown Origin: AJR Expert Panel Narrative Review. Am J Roentgenol. 2023;221:151–62. https://doi.org/10.2214/ajr.22.28726.
Mourad OPV, Detsky AS. A comprehensive evidence-based approach to fever of unknown origin. Arch Intern Med. 2003;163:545–51. https://doi.org/10.1001/archinte.163.5.545.
Hayakawa K, Ramasamy B, Chandrasekar PH. Fever of unknown origin: an evidence-based review. Am J Med Sci. 2012;344:307–16. https://doi.org/10.1097/MAJ.0b013e31824ae504.
Wu M, Wulipan F, Ma J, Qian W, Sun S, Chen P, et al. Clinical characteristics and prognostic factors of lymphoma patients initially presenting with fever of unknown origin. Am J Transl Res. 2022;14:2625–36.
Chen J, Xu D, Sun W-J, Wang W-X, Xie N-N, Ruan Q-R, et al. Differential diagnosis of lymphoma with 18F-FDG PET/CT in patients with fever of unknown origin accompanied by lymphadenopathy. J Cancer Res Clin Oncol. 2023;149:7187–96. https://doi.org/10.1007/s00432-023-04665-7.
Fiterman N, Berkman N, Kuint R. Predictors of malignant lymph node involvement in patients with mediastinal lymphadenopathy and previous cancer: A cohort study. Thorac Cancer. 2022;13:631–6. https://doi.org/10.1111/1759-7714.14311.
Wang M, Liu L, Dai Q, Jin M, Huang G. Developing a primary tumor and lymph node 18F-FDG PET/CT-clinical (TLPC) model to predict lymph node metastasis of resectable T2-4 NSCLC. J Cancer Res Clin Oncol. 2022;149:247–61. https://doi.org/10.1007/s00432-022-04545-6.
Wang W-X, Cheng Z-T, Zhu J-L, Xing M-Y, Zheng C-F, Wang S-J, et al. Combined clinical parameters improve the diagnostic efficacy of 18F-FDG PET/CT in patients with fever of unknown origin (FUO) and inflammation of unknown origin (IUO): A prospective study in China. Int J Infect Dis. 2020;93:77–83. https://doi.org/10.1016/j.ijid.2020.01.030.
Wright WF, Auwaerter PG, Dibble EH, Rowe SP, Mackowiak PA. Imaging a Fever-Redefining the Role of 2-deoxy-2-[18F]Fluoro-D-Glucose-Positron Emission Tomography/Computed Tomography in Fever of Unknown Origin Investigations. Clin Infect Dis. 2021;72:1279–86. https://doi.org/10.1093/cid/ciaa1220.
Letertre S, Fesler P, Zerkowski L, Picot M-C, Ribstein J, Guilpain P, et al. Place of the 18F-FDG-PET/CT in the Diagnostic Workup in Patients with Classical Fever of Unknown Origin (FUO). J Clin Med. 2021;10:3831. https://doi.org/10.3390/jcm10173831.
Wang Q, Li Y-M, Li Y, Hua F-C, Wang Q-S, Zhang X-L, et al. 18F-FDGPET/CT in fever of unknown origin and inflammation of unknown origin: a Chinese multi-center study. Eur J Nucl Med Mol Imaging. 2018;46:159–65. https://doi.org/10.1007/s00259-018-4121-1.
Bharucha T, Rutherford A, Skeoch S, Alavi A, Brown M, Galloway J, et al. Diagnostic yield of FDG-PET/CT in fever of unknown origin: a systematic review, meta-analysis, and Delphi exercise. Clin Radiol. 2017;72:764–71. https://doi.org/10.1016/j.crad.2017.04.014.
Takeuchi M, Nihashi T, Gafter-Gvili A, García-Gómez FJ, Andres E, Blockmans D, et al. Association of 18F-FDG PET or PET/CT results with spontaneous remission in classic fever of unknown origin. Medicine. 2018;97:e12909. https://doi.org/10.1097/md.0000000000012909.
Makis W, Ciarallo A, Gonzalez-Verdecia M, Probst S. Systemic Lupus Erythematosus Associated Pitfalls on 18F-FDG PET/CT: Reactive Follicular Hyperplasia, Kikuchi-Fujimoto Disease, Inflammation and Lymphoid Hyperplasia of the Spleen Mimicking Lymphoma. Nuclear Med Mol Imaging. 2017;52:74–9. https://doi.org/10.1007/s13139-017-0471-z.
Nawwar AA, Searle J, Hodby K, Chavda ND, Qamhia N, Lyburn ID. [18F]FDG-PET/CT Demonstration of Young Adult Onset Stills Disease Mimicking Lymphoma. BJR|case Rep. 2022;8:20210195. https://doi.org/10.1259/bjrcr.20210195.
Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RGPM, Granton P, et al. Radiomics: Extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48:441–6. https://doi.org/10.1016/j.ejca.2011.11.036.
van Rijsewijk ND, Ijpma FFA, Wouthuyzen-Bakker M, Glaudemans AWJM. Molecular Imaging of Fever of Unknown Origin: An Update. Semin Nucl Med. 2023;53:4–17. https://doi.org/10.1053/j.semnuclmed.2022.07.002.
Wright WF, Yenokyan G, Auwaerter PG. Geographic Influence Upon Noninfectious Diseases Accounting for Fever of Unknown Origin: A Systematic Review and Meta-Analysis. Open Forum Infect Dis. 2022;9:ofac396. https://doi.org/10.1093/ofid/ofac396.
Al-Zaghal A, Raynor WY, Seraj SM, Werner TJ, Alavi A. FDG-PET imaging to detect and characterize underlying causes of fever of unknown origin: an unavoidable path for the foreseeable future. Eur J Nucl Med Mol Imaging. 2018;46:2–7. https://doi.org/10.1007/s00259-018-4164-3.
Zhu W, Cao W, Zheng X, Li X, Li Y, Chen B, et al. The diagnostic value of 18F-FDG PET/CT in identifying the causes of fever of unknown origin. Clin Med. 2020;20:449–53. https://doi.org/10.7861/clinmed.2020-0268.
K M-K DK, S K-K PS. Kikuchi-Fujimoto disease as a rare cause of lymphadenopathy–two cases report and review of current literature. Otolaryngol Pol. 2013;67:1–5. https://doi.org/10.1016/j.otpol.2012.05.001.
Nair JKR, Saeed UA, McDougall CC, Sabri A, Kovacina B, Raidu BVS, et al. Radiogenomic Models Using Machine Learning Techniques to Predict EGFR Mutations in Non-Small Cell Lung Cancer. Can Assoc Radiol J. 2020;72:109–19. https://doi.org/10.1177/0846537119899526.
Wang H, Li QK, Auster M, Gong G. PET and CT features differentiating infectious/inflammatory from malignant mediastinal lymphadenopathy: A correlated study with endobronchial ultrasound-guided transbronchial needle aspiration. Radiol Infect Dis. 2018;5:7–13. https://doi.org/10.1016/j.jrid.2018.01.002.

Download PDF

Reviewers agreed at journal
27 Jul, 2024
Reviewers invited by journal
08 Jul, 2024
Editor invited by journal
06 Jul, 2024
Editor assigned by journal
28 Jun, 2024
First submitted to journal
23 Jun, 2024

You are reading this latest preprint version

Distinguishing Lymphoma from Benign Lymph Node Diseases in Fever of Unknown Origin using PET/CT Radiomics

Status:

Version 1

Abstract

Figures

Background

Methods

Study cohort

Image segmentation and extraction of radiomic features

Feature selection and model construction

Model evaluation and statistical methods

Results

Clinical characteristics of patients

Feature extraction and model establishment

Model performance evaluation

Discussion

Conclusions

Abbreviations

Declarations

References

Status:

Version 1