Quantitative ultrasound radiomics analysis to evaluate lymph nodes in patients with cancer: a systematic review

Keywords

ultrasound - cancer - quantitative analysis - lymph nodes - precision medicine

Introduction

Lymph node (LN) enlargement could indicate lymphatic spread in the human body. Ultrasound (US) imaging can determine the size, location, and morphology of LNs and present these findings in a single image.

Progress in computer science in ultrasonography has enhanced the ability to extract information from routine clinical exams. However, these methods still cannot analyze some aspects [1]. Medical data digitalization and artificial intelligence (AI) have improved the clinician’s performance (e.g., decision-making): a large amount of data, such as genomics, proteomics and metabolomics, can be efficiently analyzed using AI techniques. This approach has been increasingly used in the medical imaging field, ultimately generating a dedicated cluster called ‘radiomics’.

Radiomics analysis is a technique used to extract quantitative medical features from medical images that are not accessible by visual inspection and are determined by the spatial distribution of signal intensities and pixel interrelationships. This technique provides a large number of features that, if opportunely filtered, can be used to create and validate models that are capable of performing a specific task (e.g., classification, response to therapy prediction, recurrence investigation). The radiomics workflow comprises: 1) ROI delineation (automatic, semi-automatic, or manual) ([Fig. 1]); 2) preprocessing of the intensity histograms within the ROI; 3) feature extraction; 4) feature selection to remove redundant and non-informative features; 5) model training/validation; and 6) performance evaluation (or testing) [1]. The algorithms provide quantification of textural information, intensity, and shape and can be applied to a great variety of imaging tools, such as MRI, CT, ultrasonography, and PET. The main advantage is that information on the entire lesion can be retrieved in a noninvasive and preoperative manner and repeated multiple times.

Fig. 1 Example of US imaging and ROI delineation.

Relatively few studies have investigated the use of radiomics in US imaging, primarily because of its operator dependency (which influences reproducibility) and methodological approach. Indeed, there is a lack of a unified methodology regarding this clinical practice, and also, experience levels can be very different between physicians [1].

This systematic review aimed to evaluate the role of US-radiomics analysis in assessing lymphadenopathy in patients with cancer and the ability of radiomics to predict metastatic LN involvement in different types of cancers.

Research Method

A comprehensive literature search was carried out by one of the authors up to January 2, 2024, using the following databases: MEDLINE/PubMed, Embase, and the Cochrane Central Register of Controlled Trials. Key search terms were “lymph node”, “ultrasonography”, “texture analysis”, and “radiomics”. The search method is described in Appendix 1. Studies were selected based on the following criteria: articles in English; studies approved by an ethics committee or an institutional review committee; reviews and systematic reviews were included. Articles that did not consider LNs or US specifically, animal studies, meta-analyses, case reports, comments, letters and studies that did not provide sufficient data on the topic were excluded. No temporal restrictions were applied. No filter was applied concerning gender and age. Duplicates were identified by careful screening of the title and abstract. Full articles were retrieved when the abstract was considered relevant. The PRISMA flowchart (Supplementary Fig. 1) summarizes the search strategy adopted in this study.

Results

All 47 studies are summarized in Supplementary Table 1 [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] and were divided based on the tumor of origin: breast cancer, head and neck cancer, and cervical cancer. Most of the papers were published between 2019 and 2022 (n=19 articles), and the analyses were conducted using MATLAB, R software, SPSS, OriginPro, a papillary thyroid carcinoma (PTC), cervical LN metastasis (LNM) prediction system (custom-built, n=3), MIPAV, and Mazda. Feature extraction was mainly performed with PyRadiomics, ITK-SNAP, LifeX, MedCalc, ImageJ, PASW and Prism. The values of the area under the curve (AUC), sensitivity, specificity, and accuracy are summarized in [Table 1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48].

Table 1 Summary of the chosen methods and resulting area under the curve (AUC), sensitivity (SEN), specificity (SPE), and accuracy (ACC) for each study.

Ref.	First author	Indication	Method	AUC	SEN	SPE	ACC
ABUS, automated breast ultrasound; ACC, accuracy; ALN, axillary lymph node; AUC, area under the curve; BC, breast cancer; BMUS, B-mode ultrasound; CC, cervical cancer; CEUS: contrast-enhanced ultrasound; Clin-RS, clinical features plus radiomics scores; CT, computed tomography; deep learning radiomics; DL, deep learning; DLR, deep learning radiomics, DLRN: deep learning radiomics nomogram; DLRS, deep learning radiomics signature; EIBC, early-stage invasive breast cancer; EPC: encapsuled papillary carcinoma; ESBC: early-stage breast cancer; FNA, fine-needle aspiration; GLCM, gray-level co-occurrence matrix; GLRLM: gray-level run-length matrix; K-NN, k-nearest neighbor; LASSO, least absolute shrinkage and selection operator; LN, lymph node; LNM, lymph node metastasis; ML, metastatic lymph node; MSKCC, Memorial Sloan–Kettering Cancer Center; NA, not applicable; NSLN, non-sentinel lymph node metastasis; prRLN: posterior to the right recurrent laryngeal nerve; PTC, papillary thyroid carcinoma; PTMC, papillary thyroid microcarcinoma; QUS: quantitative ultrasound spectroscopic; SCC, squamous cell carcinoma; SEN, sensitivity; SHAP, SHapley Additive exPlanations; SLN, Sentinel lymph node; SPE, specificity; SVM; support vector machine; SWE, shear-wave elastography; TNBC, triple-negative breast carcinomas; TOT: texture of texture US, ultrasound.
Breast Cancer
[4]	Qiu et al.	Prediction of ALN metastasis	Radiomics nomogram via wavelet transform	Training: 0.816 Validation: 0.759	NA	NA	NA
[5]	Yu et al.	Prediction of ALN metastasis in EIBC	Incorporation of clinical factors with radiomics signature	Training: 0.840 Validation: 0.810	0.722	0.686	0.707
[6]	Gao et al.	Prediction of ALN metastasis	Radiomic features plus clinical characteristics	Training: 0.846 Validation: 0.733	0.667	0.901	0.833
[7]	Lee et al.	Prediction of ALN metastasis	Combination of radiomics and clinicopathological model	Training: 0.858 Validation: 0.810	NA	NA	NA
[8]	Zheng et al.	Prediction of ALN status	DL-based approach with convolutional layers technique	Prediction: 0.902 Low vs. heavy: 0.905	0.950	0.870	0.880
[9]	Sun et al.	Prediction of ALN metastasis	Three ‘image only’ radiomics models combined with molecular subtype information	Overall: 0.950	Training: 0.926 Test: 0.857	Training: 0.941 Test: 0.907	Training: 0.937 Test: 0.893
[9]	Jiang et al.	Assessment of ALN status in EIBC	Join the SWE and BMUS examinations	Training: 0.842 Validation: 0.822	NA	NA	NA
[13]	Zhang et al.	Prediction of ALN metastatic load of breast cancer	DLRN model based on preoperative ultrasound DLRS and the maximum tumor diameter of patients with BC	Training: 0.900 Test: 0.821	Training: 0.833 Test: 0.778	Training: 0.815 Test: 0.800	Training: 0.821 Test: 0.792
[14]	Chen et al.	Prediction of ALN tumor burden in patients with ESBC	Incorporating the radiomics score, ABUS imaging features and clinicopathologic features, presented with a radiomics nomogram	Training: 0.876 Test: 0.851	Training: 0.853 Test: 0.765	Training: 0.826 Test: 0.825	Training: 0.831 Test: 0.814
[35]	Li et al.	Develop a nomogram to evaluate ALN status in patients with ESBC pre-op	ABUS and other clinicopathological features are used to build a Rad-score	Training: 0.924 Validation:0.812	NA	NA	NA
[36]	Zhang et al.	Prediction of NPBC response to presurgical chemotherapy	Radiomics nomogram with clinical factors	Training: 0.855 Validation: 0.882	Training: 0.907 Validation: 0.731	Training: 0.629 Validation: 0.862	Training: 0.842 Validation: 0.778
[41]	Gu et al.	Prediction of tumor and ALN status after neoadjuvant chemotherapy	Two DL radiomics nomograms	Sys1: Validation: 0.903 Test: 0.896 Sys2: Validation: 0.853 Test: 0.863	Sys1: Test: 0.750 Sys2: NA	Sys1: NA Sys2: Test: 0.818	NA
[11]	Guo et al.	Identification of the risk of axillary NSLN	Two DL radiomics models	NSLN Training: 0.909 Test: 0.846	SLN: 0.984 NSLN: 0.984	SLN: 0.476 NSLN: 0.399	SLN: 0.801 NSLN: 0.802
[12]	Zha et al.	Accuracy of MSKCC nomogram to predict SLN status	ML-assisted analysis with MSKCC nomogram	Training: 0.901 Validation: 0.833	Training: 0.848 Validation: 0.844	Training: 0.748 Validation: 0.725	NA
[47]	Wang et al.	Prediction of SLN metastasis	DL elastography	Validation: 0.879 Testing: 0.876	Validation: 0.916 Testing: 0.884	Validation: 0.828 Testing: 0.823	Validation: 0.861 Testing: 0.850
[43]	Tang et al.	Prediction of LNM in ALN	Radiomic signature with Spearman and LASSO-selected features	Training: 0.929 Validation: 0.919	Training: 0.844 Validation: 0.727	Training: 0.892 Validation: 1.000	Training: 0.857 Validation: 0.800
[44]	Yao et al.	Prediction of SLN in EIBC	SVM + clinicopathological	Testing: 0.934	0.867	0.899	0.910
[45]	Zhou et al.	Prediction of ALN as an invasive component of EPC	Random under sampling boost machine learning model	Training: 0.867 Test: 0.875	Training: 0.917 Test: 0.857	Training: 0.731 Test: 0.806	Training: 0.766 Test: 0.814
Head and Neck Cancer
[15]	Park et al.	Prediction of lateral LNM in PTC	New texture analysis algorithm	Training: 0.710 Validation: 0.621	0.795	0.875	0.830
[16]	Kim et al.	Prediction of LNM in patients with PTMC	Automatic extraction and calculation of histogram parameters	NA	NA	NA	NA
[17]	Zhou et al.	Prediction of central LNM for PTC	Unification of US radiomics, biochemical results, and US findings in a single nomogram	Overall: 0.858	0.816	0.810	0.812
[18]	Jiang et al.	Staging of cervical LN for PTC	Incorporation of SWE and BMUS	Training: 0.851 Validation: 0.832	Training: 0.807 Validation: 0.868	Training: 0.875 Validation: 0.730	Training: 0.790 Validation: 0.789
[19]	Liu et al.	Prediction of the LN status for PTC	Combination of preoperative, demographic and pathological features	Overall: 0.90	0.770	0.880	0.850
[20]	Liu et al.	Estimation of LNM of PTC	Application of BMUS and SE combined	Overall: 0.727	0.656	0.745	0.712
[21]	Tong et al.	Prediction of central and lateral cervical LNM in PTC	Two radiomics signatures to incorporate US radiomics and clinical risk factors	Sys1: Training: 0.875 Validation: 0.870, 0.856, 0.870 Sys2: Training: 0.938 Validation: 0.881, 0.903	Sys1: Training: 0.837 Validation: 0.830, 0.771, 0.714 Sys2: Training: 0.815 Validation: 0.867, 0.733, 0.789	Sys1: Training: 0.816 Validation: 0.823, 0.851, 0.879 Sys2: Training: 0.941 Validation: 0.922, 0.930, 0.895	Sys1: Training: 0.823 Validation: 0.825, 0.813, 0.827 Sys2: Training: 0.930 Validation: 0.916, 0.910, 0.880
[22]	Tong et al.	Prediction of lateral cervical LNM in PTC	Integration of radiomic signature, US and CT reported LN status	Training: 0.946 Validation: 0.914	Training: 0.873 Validation: 0.806	Training: 0.932 Validation: 0.945	Training: 0.927 Validation: 0.930
[48]	Shen et al.	Presurgical prediction in LN-prRLNs in PTC	Radiomics + clinical model	0.849	0.964	0.330	NA
[49]	Wei et al.	Anticipation of pre-operative central LNM in PTC	BMUS + CEUS imaging and clinical features	Training: 0.891 Validation: 0.870	NA	NA	NA
[23]	Tong et al.	Prediction of cervical LNM in PTC	Unification of tumor phenotypes with genomic data	Training: 0.873 Validation: 0.831	Training: 0.836 Validation: 0.744	Training: 0.808 Validation: 0.851	Training: 0.822 Validation: 0.800
[50]	Yan et al.	Prediction of central LNM in PTC	US radiomics + morphological features	Training: 0.956 Validation: 0.973	Training: 0.922 Validation: 0.933	Training: 0.835 Validation: 0.895	Training: 0.882 Validation: 0.916
[24]	Ardakani et al.	Distinguishing of metastasis from T-F cervical LN in PTC	Feature extraction via wavelet transform compared with FNA cytology	Overall: 0.971	Training: 0.971 Validation: 0.933	Training: 0.986 Validation: 0.967	Training: 0.978 Validation: 0.950
[25]	Ardakani et al.	Detection of changes in images of LNM in PTC	Raw features extraction from histograms, transformed into low dimensional spaces	Overall: 0.996	0.993	0.985	0.989
[39]	Chung et al.	Prediction of LNM using images of PTC >1cm	GLCM and GLRLM textural features in four directions	Training: 0.687 Validation: 0.650	NA	NA	NA
[37]	Wen et al.	Prediction of central cervical LNM in patients with PTC	Predictive model based on the US radiomics signature and clinical and ultrasonic risk factors	0.600	0.200	1.000	NA
[34]	Shi et al.	Prediction of CC-LNM status in PTC	Radiomic-based XGBoost model using SHAP	0.910	0.757	0.917	0.873
[38]	Jin et al.	Prediction of central LNM in PTC patients with Hashimoto’s thyroiditis	Building of a Clin-RS model to diagnose CLNM	Training: 0.833 Validation: 0.751	NA	NA	NA
[43]	Wang et al.	Prediction of BRAF V600E mutations in PTC through LN imaging	Combination of elasticity US and grayscale US imaging	Training: 0.985 Test: 0.938	Training: 0.964 Test: 0.833	Training: 0.976 Test: 1	Training: 0.969 Test: 0.905
[40]	Xue et al.	Prediction of LNM in CC of PTC	Multimodal clinical, US features, and Rad-score to form a joint prediction model	0.934	0.951	0.837	NA
[26]	Li et al.	Prediction of LNM in thyroid cancer	Textural radiomics and tumor characteristics	Training: 0.759 Validation: 0.803	Training: 0.900 Validation: 0.727	Training: 0.860 Validation: 0.800	NA
[27]	Kwon et al.	Prediction of distant metastasis in FTC	Grayscale US-based radiomics	Overall: 0.930	Training: 0.992 Test: 0.800	Training: 0.870 Test: 0.870	Training: 0.880 Test: 0.850
[46]	Safakish et al.	Prediction of treatment outcomes through bulky LN imaging	QUS parametric maps + radiomics features + TOT to train SVM	Overall: 0.850	0.851	0.800	0.833
[31]	Dasgupta et al.	Prediction of recurrence from LN-positive imaging in SCC	Features fed to three ML classifiers	Overall: 0.740	0.760	0.710	0.750
[32]	Fatima et al.	Prediction of recurrence from LN-positive imaging in SCC	Ultrasound delta-radiomics during radiotherapy, generating an ML model	Week 1: 0.750 Week 4: 0.810	Week 1: 0.790 Week 4: 0.820	Week 1: 0.800 Week 4: 0.820	Week 1: 0.800 Week 4: 0.820
[33]	Tran et al.	Monitoring of treatment responses in carcinoma patients through LN imaging	Delta radiomics features classified with K-NN and Naive-Bayes	Naïve-Bayes: 24 h: 0.67 Week 1: 0.77 Week 4: 0.79	Naïve-Bayes: 24 h: 0.770 Week 1: 0.850 Week 4: 0.840	Naïve-Bayes: 24 h: 0.830 Week 1: 0.860 Week 4: 0.850	Naïve-Bayes: 24 h: 0.800 Week 1: 0.860 Week 4: 0.850
Cervical Cancer
[28]	Teng et al.	Prediction of LN status for CC	Comparison of automatic and manual segmentation	Sys1: 0.755 Sys2: 0.696 Sys3: 0.689 Sys4: 0.710	NA	NA	NA
[29]	Liu et al.	Classification of the etiology of cervical lymphadenopathy	Development of a new method for customizing the LASSO-SVM model	LN tuberculosis: 0.673 Cervical lymphoma: 0.623	LN tuberculosis: 0.600 Cervical lymphoma: 0.328	LN tuberculosis: 0.763 Cervical lymphoma: 0.931	LN Tuberculosis: 0.716 Cervical lymphoma: 0.817
[30]	Jin et al.	Prediction of LN status for CC	Textural radiomics algorithms	Training: 0.790 Validation: 0.770	Training: 0.670 Validation: 0.680	Training: 0.830 Validation: 0.870	NA

Breast Cancer

The vast majority of breast cancer-related studies used US-based radiomics analysis, either alone or combined with clinicopathological characteristics, to predict axillary lymph node (ALN) secondary involvement (n=13 articles) [2] [3] [4] [5] [6] [7] [8] [11] [12] [33] [39] [41] [43]. Interestingly Zheng et al. [6], Sun et al. [7], and Tang et al. [41] exploited the advantages of deep learning radiomics (DLR), reaching excellent results in predicting ALN status. Chen et al. [12] and Li et al. [33] focused on automated breast US-based analysis, building nomograms that incorporated the Rad-score with the retraction phenomenon and showing fine discriminatory power. Lastly, Zhang et al. [34] and Gu et al. [39] developed radiomics nomograms to predict treatment response and LNM status after presurgical neoadjuvant chemotherapy, obtaining good performance scores. Only one author [43] investigated the prediction of ALN metastases as an invasive component of encapsulated papillary carcinoma.

Overall, the ALN metastasis prediction studies show promising results, under consideration of validation/testing values: all the studies were successful in terms of AUC values. In fact, the lower AUC value is reported by Gao et al. [4] at 0.733, while eight studies report an AUC value over 0.800 [3] [5] [8] [11] [12] [33] [39] [43] and three over 0.900 [6] [7] [11].

Finally, Guo et al. [9], Zha et al. [10], Yao et al. [42], and Wang et al. [45] focused on sentinel lymph node prediction, an interesting topic in metastatic development. In this case, the AUC values are all above 0.800, with a notable value of 0.934 for Yao et al. [42].

Head and Neck Cancer

Head and neck cancer studies mainly focused on thyroid cancer, specifically PTC and head and neck squamous cell carcinoma (HNSCC). We identified 20 studies investigating LN status in patients with PTC [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [32] [35] [36] [37] [38] [46] [47] [48], one study on follicular thyroid carcinoma [25], one on treatment outcome prediction in head and neck malignancies [44], and three studies on HNSCC [29] [30] [31].

Most studies (n=15 articles) on PTC showed that it is possible to predict lateral or central LNM using radiomics [13], although one study from Kim et al. [14] was unsuccessful. Radiomics analysis included models combining shear-wave elastography and B-mode US combined Rad-score [16], strain elastography US and B-mode US [18], radiomics and clinical features [21] [36] [37] [38] [48], textural and radiological features [22], grayscale texture analysis [23], and clinical, radiomics, and extrathyroidal extension features [35].

One study by Wang et al. [40] used a combined model of grayscale plus elasticity features to predict BRAF ^V600E mutations in PTC, obtaining good results, while Li et al. [24] focused on the role of radiomics to influence pre-interventional decisions in PTC. Safakish et al. [44] conducted an innovative study merging information from quantitative US parametric maps, radiomics features, and the novel texture-on-texture features to improve model performance.

The overall performance in the PTC context is heterogeneous, with one failed study, three studies that achieve 0.621, 0.600 and 0.65 AUC values, respectively [13] [35] [37], nine studies that range between 0.727 and 0.881 [15] [16] [17] [18] [19] [21] [36] [44] [46] and seven with an AUC over 0.900 [17] [20] [22] [23] [31] [38] [47].

Studies on HNSCC focused on quantitative US delta-radiomics to investigate recurrence in patients undergoing radiotherapy, obtaining good prediction results (0.740, 0.810, 0.790 AUC values, respectively) [29] [30] [31].

Cervical Cancer

We identified three studies on cervical cancer. They focused on the consequences of manual vs. automatic segmentation (AUC: 0.755) [26], the creation of a classification model to predict the etiology of cervical lymphadenopathies (AUC: 0.673) [27], and the prediction of LN status in patients with early-stage cervical cancer (AUC: 0.770) [28].

Discussion

Overall, this systematic review shows that US-radiomics is useful for evaluating lymphadenopathy and predicting LN status in patients with breast cancer, head and neck cancer, and cervical cancer. This noninvasive tool can provide information to guide clinician decision-making, determine the best treatment approach, and avoid unnecessary invasive procedures. Moreover, integrating radiomics and genomic features (radiogenomics) has revealed encouraging results, providing opportunities to improve the diagnosis and prognosis of different types of cancer [50].

Breast Cancer

The LN application concerning breast neoplasms mostly regards the status of ALN, which is essential for the diagnostic process. Several studies have shown that US-radiomics can be an accurate and noninvasive presurgical method to guide treatment decision-making and avoid unnecessary invasive surgery in patients with breast cancer. Despite some limitations related to the small sample size, the need to evaluate results on prospective populations, and the use of images obtained with different systems and radiologists, several studies have highlighted that the ability of US-based radiomics to predict ALN metastasis is higher when the nomogram also incorporates clinical data, such as age, mass size, tumor type, neoadjuvant chemotherapy, clinicopathologic and textural features, or clinical risk factors [3] [4] [5]. Another interesting option is to combine shear-wave elastography and B-mode US examinations, as showed by Jiang et al. [16], who developed and validated a peculiar nomogram that exhibited good predictive ability in discriminating low vs. heavy burden ALN metastasis in patients with early-stage breast cancer.

Other tools investigated to quantify the ALN metastasis load status include automated breast US and machine learning or deep learning technology. In cases of early-stage breast cancer, ALN preoperative investigation results are indispensable for therapeutic evaluation choices. For this reason, several studies have focused on developing new radiomics nomograms for assessing ALN metastasis burden. Specifically, a recent study has shown that the use of a radiomics nomogram (in which clinicopathologic, imaging, and radiomic automated breast US features converge) can be a valuable tool for predicting the tumor burden of ALN metastasis and has been shown to have a favorable prediction performance (Chen et al. [12]). Moreover, Li et al. [33] discovered that a high ALN metastasis burden is independently associated with automated breast US retraction phenomenon, tumor size, Rad-score, and US-reported LN status in patients with early-stage breast cancer.

Deep learning approaches are also indispensable for evaluating ALNs, and different radiomics models have been developed. Zhang et al. [11] built a DLR nomogram model combining preoperative US DLR-score and maximum tumor diameter to accurately quantify the ALN metastasis load status. Zheng et al. [6] took advantage of the “convolutional layers” technique to extract and reduce features from conventional US and shear-wave elastography to determine the right treatment option in early-stage breast cancer. Guo et al. [9] designed two DLR models to predict the presence of metastasis in sentinel lymph nodes and no sentinel lymph nodes in primary breast cancer, demonstrating good predictive performance. Sun et al. [7] built three ‘image-only’ radiomics models using the random forest algorithm, focusing on three ROI images combined with molecular subtype information, emphasizing the importance of the tumor surroundings when investigating the prediction of ALN metastasis for breast cancer. Zha et al. [10] explored whether a machine learning-assisted analysis improves the prediction of preoperative sentinel lymph node status. They found that a combined nomogram, obtained by incorporating the Memorial Sloan Kettering Cancer Center nomogram and US-based radiomics score, resulted in a better performance with respect to the Memorial Sloan Kettering Cancer Center nomogram alone. Finally, an interesting methodology is introduced by Wang et al. [45], which proposed a deep learning-elastography model with high prediction capabilities on ALN metastatic status.

Treatment options prior to operative surgeries are often a reason for clinical discussion, especially in settings like breast cancer. In this regard, radiomics can help with treatment choices by predicting response to presurgical neoadjuvant chemotherapy. Zhang et al. [34] and Gu et al. [39] explored this aspect: the former developed a radiomics nomogram with a clinical factors model that has the potential to be a good, personalized aid in clinical decision-making, while the latter developed a deep learning radiomics nomogram-PCR and a deep learning radiomics nomogram-LNM with good predictive performance with respect to the specificity of each patient.

Head and Neck Cancer

In thyroid cancer, judging the LNM is very important because it is related to local recurrence, distant metastasis, and thyroid stage, thus indicating the best surgical plan [49]. For thyroid LNs, the most used analysis program is MATLAB, while there is a certain variability in feature extraction programs, such as PyRadiomics, MedCalc, ITK-SNAP, and others (Supplementary Table 1).

In recent years, most studies have focused on using US-radiomics to predict LNM in patients with PTC, showing positive results and identifying some parameters to predict LNM evolution and patient prognosis. For example, irregular shape and microcalcification are the best markers to predict the evolution of LNM in thyroid cancer (Li et al. [24]). It is important to emphasize the critical role played by the operator, as US performance is heavily affected by the experience of the physician [13]. Notably, Kim and colleagues [14] showed that US texture analysis is not useful to discriminate metastatic LNs (higher entropy) in papillary thyroid carcinoma: one cause could be the subjective influence of the eight different radiologists who performed the staging and the manual segmentation. In our opinion, in the coming years, automatic segmentation could minimize the intercorrelation bias.

As demonstrated by several recent studies, to bypass this limitation, a multimodal approach combining US-radiomics with additional data (demographic data, clinicopathological information, genetics, or shear-wave elastography) is considered better for predicting LN status and patient risk in thyroid carcinoma [15] [16] [17] [18] [46] [47] [48]. Specifically, methods based on unifying US-radiomics with biochemical results and US findings, the radiomics nomogram that incorporates shear-wave elastography with B-mode US images, or the model combining shear-wave elastography and B-mode US have been considered superior to single-component models based on the estimation of LNM in PTC [15] [16] [17] [18]. All these approaches have the potential to facilitate early medical care and ease overdiagnosis.

Many radiomics models have been developed to predict cervical LNM status. Tong et al. [19] [20] [21] largely investigated different models of US-radiomics, determining the creation of two radiomics signatures. Initially, they merged tumor phenotypes with genomic data, successfully identifying the molecular properties of the tumors. Subsequently, they developed a novel radiomics nomogram integrating radiomics signature, US-reported, and CT-reported lateral cervical LN status to assess cervical LN status in patients with PTC undergoing either total thyroidectomy or lobectomy. Finally, they designed two radiomics signatures incorporating US-radiomics and clinical factors, obtaining significantly higher Rad-scores in the central and lateral LNM groups than the non-LNM groups.

Additional US-radiomics models include those that combine US-radiomics features and clinical variables (young age and large tumor size), incorporate the extrathyroidal extension status, and integrate clinical parameters with US imaging information, surpassing the evaluation ability of radiologists, and a clinical-multimodal US combined model [32] [35] [37] [38]. Lastly, Ardakani et al. [22] [23] developed two models to differentiate tumor-free from metastatic cervical LN in patients with PTC; the models, which combined radiologic features and texture analysis, proved useful for pattern recognition and classification.

Some studies focusing on other types of thyroid tumors have also been identified. Wang et al. [40] combined elasticity US with grayscale US to diagnose BRAF ^V600E mutations pre-surgery in PTC. The analyses were retrospectively conducted on 138 patients (± BRAF mutation) with PTC who underwent preoperative US; 479 radiomics features were extracted from the grayscale and elasticity ultra-sonograms. At the end of the analysis, eight and five radiomics features were extracted from the grayscale ultrasonogram and the elasticity ultrasonogram, respectively, determining the development of three radiomics models.

Kwon et al. [25] demonstrated how US radiomics can predict distant metastasis of follicular thyroid carcinoma, the second most common thyroid cancer. The results showed that some clinicopathologic variables (age, size, widely invasive histology, LNM, and extrathyroidal extension), US characteristics (orientation, echogenicity, rim calcification, nodule-in-nodule appearance), radiomics signature, and widely invasive histology were significantly associated with distant metastasis of follicular thyroid carcinoma.

Studies on HNSCC have also shown that quantitative US-based radiomics can be useful in predicting tumor recurrence and patient survival after RT [29] [30] [31]. Dasgupta et al. [29] published the first study reporting the ability of quantitative US-based radiomics to predict tumor behavior and its impact on survival in patients with HNSCC. Another article by Fatima et al. [30] showed that US delta-radiomics could predict recurrence in patients with HNSCC treated with RT, with an increase in accuracy over time, up to 82% at week 4 of RT. Tran et al. [31] also reported the ability of quantitative US delta-radiomics to predict response to treatment in head and neck malignancies: again, the prediction ability of the model improved with time, with prognostic accuracy of 80%, 86%, and 85%, after 24 hours, 1 week, and 4 weeks of treatment, respectively.

Cervical Cancer

In the last few years, some studies have emerged on the use of US-based radiomics in determining LN status in patients suffering from cervical cancer [26] [27] [28]. Generally, the results obtained from different and independent studies have shown that the radiomics approach based on the textural features of US images can be considered a promising noninvasive prediction method used to facilitate preoperative identification of LNs in patients with early-stage cervical cancer [30]. Furthermore, automatic segmentation in patients with early-stage cervical cancer presented better prediction performance than manual segmentation [26]. Differently, Liu et al. [27] focused on the capability of US-based radiomics to classify the etiology of cervical lymphadenopathy and to distinguish between cervical LN tuberculosis, cervical lymphoma, reactive LN hyperplasia, and metastatic LNs. The customized LASSO SVM model showed adequate and repeatable performance for multiple classification diagnoses of cervical lymphadenopathy, particularly in cervical LN tuberculosis and cervical lymphoma.

Conclusion

Experimental evidence suggests that radiomics can be a useful tool in predicting and characterizing LN masses. It is a noninvasive and effective method to expand knowledge of nonvisible features and their correlation with LN status. Most of the current evidence focuses on breast cancer, head and neck cancer, and cervical cancer. Further studies should focus on other types of cancer, such as melanoma, because of the importance of US for follow-up and because of how the decision-making process for this kind of malignancies has led to the development of new therapeutic strategies in precision medicine.

The radiomics approach has some limitations, such as its reproducibility, due to differences in instruments used, setup parameters, and software used. The lack of a defined and unified procedure (and software) for acquiring and analyzing features makes the radiomics methods incomplete for real hospital implementation. More studies with a larger number of patients (and subsequently images) are needed to improve the robustness of radiomics models.

• References

• 1 van Timmeren JE, Cester D, Tanadini-Lang S. et al. Radiomics in medical imaging-"how-to" guide and critical reflection. Insights Imaging 2020; 11: 91
• 2 Qiu X, Jiang Y, Zhao Q. et al. Could ultrasound-based radiomics noninvasively predict axillary lymph node metastasis in breast cancer?. J Ultrasound Med 2020; 39: 1897-1905
• 3 Yu FH, Wang JX, Ye XH. et al. Ultrasound-based radiomics nomogram: A potential biomarker to predict axillary lymph node metastasis in early-stage invasive breast cancer. Eur J Radiol 2019; 119: 108658
• 4 Gao Y, Luo Y, Zhao C. et al. Nomogram based on radiomics analysis of primary breast cancer ultrasound images: prediction of axillary lymph node tumor burden in patients. Eur Radiol 2021; 31: 928-937
• 5 Lee SE, Sim Y, Kim S. et al. Predictive performance of ultrasonography-based radiomics for axillary lymph node metastasis in the preoperative evaluation of breast cancer. Ultrasonography 2021; 40: 93-102
• 6 Zheng X, Yao Z, Huang Y. et al. Deep learning radiomics can predict axillary lymph node status in early-stage breast cancer. Nat Commun 2020; 11: 1236
• 7 Sun Q, Lin X, Zhao Y. et al. Deep learning vs. radiomics for predicting axillary lymph node metastasis of breast cancer using ultrasound images: don't forget the peritumoral region. Front Oncol 2020; 10: 53
• 8 Jiang M, Li CL, Luo XM. et al. Radiomics model based on shear-wave elastography in the assessment of axillary lymph node status in early-stage breast cancer. Eur Radiol 2022; 32: 2313-2325
• 9 Guo X, Liu Z, Sun C. et al. Deep learning radiomics of ultrasonography: Identifying the risk of axillary non-sentinel lymph node involvement in primary breast cancer. EBioMedicine 2020; 60: 103018
• 10 Zha HL, Zong M, Liu XP. et al. Preoperative ultrasound-based radiomics score can improve the accuracy of the Memorial Sloan Kettering Cancer Center nomogram for predicting sentinel lymph node metastasis in breast cancer. Eur J Radiol 2021; 135: 109512
• 11 Zhang H, Zhao T, Zhang S. et al. Prediction of axillary lymph node metastatic load of breast cancer based on ultrasound deep learning radiomics nomogram. Technol Cancer Res Treat 2023; 22
• 12 Chen Y, Xie Y, Li B. et al. Automated breast ultrasound (ABUS)-based radiomics nomogram: an individualized tool for predicting axillary lymph node tumor burden in patients with early breast cancer. BMC Cancer 2023; 23: 340
• 13 Park VY, Han K, Kim HJ. et al. Radiomics signature for prediction of lateral lymph node metastasis in conventional papillary thyroid carcinoma. PLoS One 2020; 15: e0227315
• 14 Kim SY, Lee E, Nam SJ. et al. Ultrasound texture analysis: Association with lymph node metastasis of papillary thyroid microcarcinoma. PLoS One 2017; 12: e0176103
• 15 Zhou SC, Liu TT, Zhou J. et al. An Ultrasound radiomics nomogram for preoperative prediction of central neck lymph node metastasis in papillary thyroid carcinoma. Front Oncol 2020; 10: 1591
• 16 Jiang M, Li C, Tang S. et al. Nomogram based on shear-wave elastography radiomics can improve preoperative cervical lymph node staging for papillary thyroid carcinoma. Thyroid 2020; 30: 885-897
• 17 Liu T, Ge X, Yu J. et al. Comparison of the application of B- mode and strain elastography ultrasound in the estimation of lymph node metastasis of papillary thyroid carcinoma based on a radiomics approach. Int J Comput Assist Radiol Surg 2018; 13: 1617-1627
• 18 Liu T, Zhou S, Yu J. et al. Prediction of lymph node metastasis in patients with papillary thyroid carcinoma: a radiomics method based on preoperative ultrasound images. Technol Cancer Res Treat 2019; 18: 1533033819831713
• 19 Tong Y, Sun P, Yong J. et al. Radiogenomic analysis of papillary thyroid carcinoma for prediction of cervical lymph node metastasis: a preliminary study. Front Oncol 2021; 11: 682998
• 20 Tong Y, Li J, Huang Y. et al. Ultrasound-Based Radiomic Nomogram for Predicting Lateral Cervical Lymph Node Metastasis in Papillary Thyroid Carcinoma. Acad Radiol 2021; 28: 1675-1684
• 21 Tong Y, Zhang J, Wei Y. et al. Ultrasound-based radiomics analysis for preoperative prediction of central and lateral cervical lymph node metastasis in papillary thyroid carcinoma: a multi-institutional study. BMC Med Imaging 2022; 22: 82
• 22 Abbasian Ardakani A, Reiazi R. et al. A clinical decision support system using ultrasound textures and radiologic features to distinguish metastasis from tumor-free cervical lymph nodes in patients with papillary thyroid carcinoma. J Ultrasound Med 2018; 37: 2527-2535
• 23 Ardakani AA, Rasekhi A, Mohammadi A. et al. Differentiation between metastatic and tumour-free cervical lymph nodes in patients with papillary thyroid carcinoma by grey-scale sonographic texture analysis. Pol J Radiol 2018; 83: e37-e46
• 24 Li F, Pan D, He Y. et al. Using ultrasound features and radiomics analysis to predict lymph node metastasis in patients with thyroid cancer. BMC Surg 2020; 20: 315
• 25 Kwon MR, Shin JH, Park H. et al. Radiomics based on thyroid ultrasound can predict distant metastasis of follicular thyroid carcinoma. J Clin Med 2020; 9: 2156
• 26 Teng Y, Ai Y, Liang T. et al. the effects of automatic segmentations on preoperative lymph node status prediction models with ultrasound radiomics for patients with early stage cervical cancer. Technol Cancer Res Treat 2022; 21
• 27 Liu Y, Chen J, Zhang C. et al. Ultrasound-based radiomics can classify the etiology of cervical lymphadenopathy: a multi-center retrospective study. FrontOncol 2022; 12: 856605
• 28 Jin X, Ai Y, Zhang J. et al. Non-invasive prediction of lymph node status for patients with early-stage cervical cancer based on radiomics features from ultrasound images. Eur Radiol 2020; 30: 4117-4124
• 29 Dasgupta A, Fatima K, DiCenzo D. et al. Quantitative ultrasound radiomics in predicting recurrence for patients with node-positive head-neck squamous cell carcinoma treated with radical radiotherapy. Cancer Med 2021; 10: 2579-2589
• 30 Fatima K, Dasgupta A, DiCenzo D. et al. Ultrasound delta-radiomics during radiotherapy to predict recurrence in patients with head and neck squamous cell carcinoma. Clin Transl Radiat Oncol 2021; 28: 62-70
• 31 Tran WT, Suraweera H, Quiaoit K. et al. Quantitative ultrasound delta-radiomics during radiotherapy for monitoring treatment responses in head and neck malignancies. Future Sci OA 2020; 6: FSO624
• 32 Shi Y, Zou Y, Liu J. et al. Ultrasound-based radiomics XGBoost model to assess the risk of central cervical lymph node metastasis in patients with papillary thyroid carcinoma: Individual application of SHAP. Front Oncol 2022; 12: 897596
• 33 Li N, Song C, Huang X. et al. Optimized radiomics nomogram based on automated breast ultrasound system: a potential tool for preoperative prediction of metastatic lymph node burden in breast cancer. Breast Cancer (Dove Med Press) 2023; 15: 121-132
• 34 Zhang H, Cao W, Liu L. et al. Non-invasive prediction of node-positive breast cancer response to presurgical neoadjuvant chemotherapy therapy based on machine learning of axillary lymph node ultrasound. J Transl Med 2023; 21: 337
• 35 Wen Q, Wang Z, Traverso A. et al. A radiomics nomogram for the ultrasound-based evaluation of central cervical lymph node metastasis in papillary thyroid carcinoma. Front Endocrinol (Lausanne) 2022; 13: 1064434
• 36 Jin P, Chen J, Dong Y. et al. Ultrasound-based radiomics nomogram combined with clinical features for the prediction of central lymph node metastasis in papillary thyroid carcinoma patients with Hashimoto's thyroiditis. Front Endocrinol (Lausanne) 2022; 13: 993564
• 37 Chung HJ, Han K, Lee E. et al. radiomics analysis of grayscale ultrasonographic images of papillary thyroid carcinoma > 1 cm: potential biomarker for the prediction of lymph node metastasis. J Korean Soc Radiol 2023; 84: 185-196
• 38 Xue J, Li S, Qu N. et al. Value of clinical features combined with multimodal ultrasound in predicting lymph node metastasis in cervical central area of papillary thyroid carcinoma. J Clin Ultrasound 2023; 51: 908-918
• 39 Gu J, Tong T, Xu D. et al. Deep learning radiomics of ultrasonography for comprehensively predicting tumor and axillary lymph node status after neoadjuvant chemotherapy in breast cancer patients: A multicenter study. Cancer 2023; 129: 356-366
• 40 Wang YG, Xu FJ, Agyekum EA. et al. Radiomic model for determining the value of elasticity and grayscale ultrasound diagnoses for predicting BRAFV600E mutations in papillary thyroid carcinoma. Front Endocrinol (Lausanne) 2022; 13: 872153
• 41 Tang YL, Wang B, Ou-Yang T. et al. Ultrasound radiomics based on axillary lymph nodes images for predicting lymph node metastasis in breast cancer. Front Oncol 2023; 13: 1217309
• 42 Yao J, Zhou W, Xu S. et al. Machine learning-based breast tumor ultrasound radiomics for pre-operative prediction of axillary sentinel lymph node metastasis burden in early-stage invasive breast cancer. Ultrasound Med Biol 2024; 50: 229-236
• 43 Zhou J, Liu C, Shi Z. et al. Application of ultrasound-based radiomics models of breast masses to predict invasive components of encapsulated papillary carcinoma. Quant Imaging Med Surg 2023; 13: 6887-6898
• 44 Safakish A, Sannachi L, DiCenzo D. et al. Predicting head and neck cancer treatment outcomes with pre-treatment quantitative ultrasound texture features and optimising machine learning classifiers with texture-of-texture features. Front Oncol 2023; 13: 1258970
• 45 Wang C, Zhao Y, Wan M. et al. Prediction of sentinel lymph node metastasis in breast cancer by using deep learning radiomics based on ultrasound images. Medicine (Baltimore) 2023; 102: e35868
• 46 Shen B, Zhou C, Xu C. et al. Ultrasound-based radiomics for predicting metastasis in the lymph nodes posterior to the right recurrent laryngeal nerve in patients with papillary thyroid cancer. Curr Med Imaging 2023;
• 47 Yan X, Mou X, Yang Y. et al. Predicting central lymph node metastasis in patients with papillary thyroid carcinoma based on ultrasound radiomic and morphological features analysis. BMC Med Imaging 2023; 23: 111
• 48 Wei T, Wei W, Ma Q. et al. Development of a clinical-radiomics nomogram that used contrast-enhanced ultrasound images to anticipate the occurrence of preoperative cervical lymph node metastasis in papillary thyroid carcinoma patients. Int J Gen Med 2023; 16: 3921-3932
• 49 Maksimovic S, Jakovljevic B, Gojkovic Z. Lymph node metastases papillary thyroid carcinoma and their importance in recurrence of disease. Med Arch 2018; 72: 108-111
• 50 Liu Z, Duan T, Zhang Y. et al. Radiogenomics: a key component of precision cancer medicine. Br J Cancer 2023; 129: 741-753