Clinical characteristics of children with primary tracheobronchial tumors and SVM model in mucoepidermoid carcinoma Prediction

Objective To determine the clinical characteristics, treatment, and prognosis of primary tracheobronchial tumors (PTT) in children. The medical records of through were reviewed retrospectively. The clinical features, imaging, and bronchoscopic manifestations, treatment, and outcomes of these patients were summarized. Sixteen children were hospitalized with PTT during the study period: ve (31.3%) with mucoepidermoid carcinoma (ME), three (18.8%) with inammatory myobroblastic tumors (IMT), two (12.5%) with sarcoma, two (12.5%) with papillomatosis, and one (6.3%) each with carcinoid carcinoma, adenoid cystic carcinoma (ACC), hemangioma, and and we used some and learning decision The at diagnosis of were progressive dyspnea (37.5%), hemoptysis and fever (25%). The most frequent complication on chest imaging was atelectasis (56.3%), followed by bronchiectasis and emphysema (12.5%). Of the 16 patients, seven underwent surgical treatment, eight underwent bronchoscopic tumor resection, and one died. Of the 11 children followed-up, three experienced recurrence, including two who underwent subsequent tracheotomies, both of whom have varying degrees of post-activity shortness of breath. The other eight patients did not experience recurrence. No deaths were observed during follow-up. model showed high accuracy with respect to identifying ME. Chest CT and bronchoscopy can diagnose PTT effectively. Surgery is the treatment of choice, although bronchoscopic intervention can achieve good results in patients with unresectable tumors. The prognosis of the 11 followed-up children was good.


Introduction
Primary tracheobronchial tumors (PTT) are extremely rare in children and are misdiagnosed easily. Few reports describe patients with these tumors, making their clinical presentation, diagnosis, treatment, and outcome unclear. This study retrospectively reviewed the clinical data of children who were hospitalized with pathologically con rmed PTT in the Children's Hospital of Chongqing Medical University from January 1995 to January 2020, and then followed-up by telephone in July 2020.

Materials And Methods
The medical records of all children diagnosed with PTT from January 1995 to January 2020 at the respiratory department, thoracic surgery ward, and ear, nose, and throat department of the Children's Hospital of Chongqing Medical University were reviewed retrospectively. Data collected included demographic characteristics, clinical presentation, results of chest X-rays and computed tomography (CT) scans, pathological diagnosis, lesion location, treatment, followup, and clinical outcome. Patients were included if they were under 18 years old and had an intracavitary lesion of the trachea or large bronchus con rmed by chest CT and bronchoscopy, with pathology con rmed by biopsy 1,2 . Patients with lung cancer, laryngeal cancer, and metastatic tracheobronchial tumors were excluded. At the same time, the clinical manifestations and imaging characteristics collected were used to establish a database to construct an identi cation model for the most common tracheobronchial tumor type, ME. We use Gaussian naïve Bayes model, SVM(support vector machine) model, and decision tree model to build the model respectively.

Radiological examination
Most of these children underwent chest CT with enhancement and airway reconstruction, with chest CT usually showing bronchial luminal obstruction in the lesion, with soft tissue shadowing, with or without bronchiectasis, atelectasis, or compensatory emphysema (Fig. 1). The most common abnormalities were atelectasis (9/16, 56.3%), followed by bronchiectasis (4/16, 25%) and emphysema (2/16, 12.5%). Maximum tumor diameter ranged from 5 to 40 mm. Seven tumors were located in the left main bronchus, three in the right main bronchus, and six in the trachea. PET/CT scanning of one patient with ME (moderately malignant) and carcinoid carcinoma revealed an SUV in the primary lesion that was signi cantly higher than that in the surrounding tissue, which is consistent with manifestations of a malignant tumor. F-18 FDG accumulation in the lesion has been reported in high-and low-grade MEs 3 .

Diagnosis and treatment
Fifteen children underwent bronchoscopy and biopsy. Seven were treated surgically. Three (19%) underwent sleeve lobectomies: two underwent left upper sleeve lobectomies, and one underwent a right middle sleeve lobectomy. In addition, one (6%) patient underwent a right middle and lower lobectomy, and three (19%) underwent tracheal sleeve lobectomies with end-to-end anastomosis, including one who also underwent partial thyroidectomy. The remaining eight children underwent local tumor resection via bronchoscopy: ve (31%) underwent tumor resection alone, and two (13%) underwent tumor resection followed by chemotherapy. Surgery was successful in all 15 patients.
One child died during hospitalization due to sudden onset of extreme dyspnea. Bronchoscopy revealed complete obstruction of the left main bronchus and two-thirds of the right bronchus by the tumor. An autopsy con rmed the diagnosis of rhabdomyosarcoma.

Follow-up and outcome
Of the 15 children, two were lost to follow-up after 5 and 60 months, and four were never followed-up. The remaining nine children were reached successfully by telephone in July 2020. The average follow-up duration was 46 months (range, 5-182 months). One child (patient no. 3) experienced tumor recurrence within 2 months after tracheal sleeve resection; this patient underwent local tumor resection twice, with no recurrence for 18 months thereafter. However, this patient still experiences wheezing after strenuous activity. Two children with intratracheal papillomatosis (patient nos. 13 and 14) experienced tumor recurrence 6-8 times and underwent repeated local tumor resection, with no new recurrence at the last follow-up. One still experiences shortness of breath after activity, whereas the other can participate in normal activities; both have hoarseness. Two patients (patient nos. 2 and 4) who underwent local tumor resection plus chemotherapy are still receiving chemotherapy, with patient number 2 being treated with ifosfamide, etoposide, vindesine, methotrexate, and prednisone; their activities are not affected. The remaining children show no evidence of disease recurrence. No deaths were observed during the follow-up period.
3.6 Recognition model of ME ME carcinoma was the most common tracheobronchial tumor in children. A total of 16 characteristic parameters were selected to construct a database: ten clinical manifestations (cough, fever, progressive dyspnea, weight loss, wheezing, dysphagia, breathlessness, hemoptysis, chest pain, and hoarseness); three tumor locations (left bronchus, right bronchus, and trachea); and three imaging features (atelectasis, bronchiectasis, and emphysema). Then, a Gaussian naïve Bayes model, an SVM model, and a decision tree model were used to predict incidence of mucoepidermoid carcinoma (Table 4). All cases were assigned randomly to a training set (11 cases) and a test set ( ve cases). The training set was used to select characteristics and to construct recognition models, and the test set was used to evaluate the prediction performance of the models. The area under the ROC curve (AUC), accuracy, and F1-score were used to evaluate the models. We found that the SVM model showed the best performance, followed by the decision trees and the Gaussian plain Bayes model. The SVM model uses python sklearn, and the kernel is linear; the relaxation coe cient is 6.13, and the other parameters are default settings. The accuracy, AUC, and F1-score of the test set were all 1 (Fig. 2), which indicates that the model shows good predictive e ciency in the test set.

Discussion
PTTs are very rare in children. Here, we found that the incidence of tumors in pre-school or school-age children was higher in boys than in girls, and that the most common tumor was ME (31.3%). Different from the most common types of tumor reported abroad 5 , ME carcinoma is the most common type of PTT reported in domestic studies, even accounting for up to 74% of cases 4 . Thus, ME is a malignant tumor with a high incidence rate, meaning that identi cation is important. The SVM model showed high accuracy with respect to identifying ME. Therefore, in the absence of invasive pathological biopsy, the model can be used to analyze clinical data from patients with tracheobronchial tumors and can help to identify the typeof the tumor; this will be especially useful for young doctors or doctors in primary hospitals as it provides an important reference for follow-up treatment and clinical research. However, in view of the small number of cases in our group, the model shows some instability and requires further veri cation using other independent data.
No speci c respiratory symptoms are present during the early stage of the PPTs 6 . In our cohort, the most common symptom was a cough, followed by breathlessness, wheezing, progressive dyspnea, and fever. The incidence of speci c symptoms such as hemoptysis and chest pain is similar in children and adults 4 , whereas non-speci c symptoms such as fever, cough, and wheezing are more common in children. X-rays of the 16 children identi ed lung infection in seven. The remaining nine were negative, resulting in misdiagnoses that included pneumonia, asthma, or foreign bodies, with one child being misdiagnosed with tuberculosis for 5 years. Finally, these children were diagnosed with PTT by chest CT and bronchoscopy. The rate of misdiagnosis of PTTs has decreased gradually over time, as has the time from appearance of symptoms to a correct diagnosis; this is due to greater use of chest CT and bronchoscopy. Chest CT can assess the size and location of the lesion, the extent of in ltration into surrounding tissues, the enlargement of mediastinal lymph nodes, and the involvement of the lung parenchyma 6,7 . The high calci cation rate (50%) seen in ME 8 can be con rmed by CT. Bronchoscopy can visualize the site, shape, and size of the lesion, enabling biopsies to determine the nature of the lesion; however, the positivity rate of bronchial lavage is low 9 .
Surgery is the treatment of choice for PTT in children, with sleeve lobectomy often chosen to preserve lung function [4][5][6] . Complete surgical resection is effective in patients with malignant tumors 10 ; however, the bene ts of adjuvant radiotherapy and chemotherapy are unclear, so they are used rarely.
Chemotherapy and radiotherapy are still recommended for children with tumor recurrence 7 , and radiotherapy is a viable option for patients with highly malignant tumors with positive surgical margins 11 . Studies of targeted therapy for PTT show that recurrent respiratory papillomatosis responds well to bevacizumab 12 . In addition, adjuvant treatment with an ALK inhibitor such as cyclozolidinib improves the prognosis of children with IMT and partial ALK gene rearrangement [13][14][15] .
In addition to surgical resection, bronchoscopic intervention is effective in treating patients with PTT 9 . However, the bene ts of endoscopic local resection of tumors are unclear because endoscopic resection has several limitations, including a high risk of transmural injury, an inability to coagulate larger diameter vessels, and a risk of airway tree dislocation 16 . CT angiography before brilloscopic intervention showed that two patients in the present cohort had large vessels deep within the tumor. After consultation with a cardiologist to reduce the risk of bleeding, both patients underwent minimally invasive vessel closure. In addition to the higher intraoperative risk, the recurrence rate was higher after local resection than after surgical treatment. Indeed, 43% of patients who underwent local resection experienced tumor recurrence during follow-up period 4 . Because some of these patients later required tracheal cricotomy, single local tumor resection through bronchoscopy is not recommended. At present, bronchoscopy is used mainly to assess the degree of airway narrowing, to take biopsies to con rm the diagnosis, and to perform temporary palliative treatment such as deburring, lavage, or APC plus cryotherapy 7 in patients with unresectable lesions, both to reduce airway obstruction and improve patient quality of life.
Factors affecting the prognosis of children with PTT include histologic classi cation, pathologic grade, tumor size, lymph node status, distant metastasis, and the extent and ability of surgery [17][18][19] , with pathologic type being the most important factor 19 . For example, ME (the most common type of PTT) can be classi ed as low, intermediate, and high-grade tumors 20,21 . The Adult Tracheobronchial ME Study reported that the 5 year survival rate of patients with lowgrade ME was 57%, whereas the survival rate of patients with high-grade ME was signi cantly lower 21,22 , with a 1 year survival rate of only 20% 18 . Moreover, survival rates differed signi cantly in patients who did and did not undergo surgery. By contrast, the 5 year survival rate of children with ME is estimated to be 95%; no deaths were reported in children with low-to intermediate-grade ME, with those having high-grade ME or distant metastases having a lower survival rate. The presence of a t(11;19) (q21;p13) translocation is also associated with a good prognosis in children 18 . Chemotherapy is associated with poorer late health and lower long-term survival rates 23 . Except for one child in our cohort who was diagnosed with sarcoma and died during hospitalization, the prognosis of the 11 children who were followed-up was good. Three patients experienced tumor recurrence, with two undergoing subsequent tracheotomy. Their prognosis, however, remained good after a maximum follow-up of 182 months. The other eight patients, including three treated surgically and ve treated by bronchial intervention, did not experience tumor recurrence after a maximum follow-up of 76 months. Guidelines suggest that bronchoscopy should be performed every 6 months for the rst year and then annually thereafter 24 .
In summary, pediatric PTT is relatively rare, with an insidious onset and an uncharacteristic clinical presentation 6,7 . Increased tumor volume may cause tracheal and bronchial lumen narrowing and, when combined with respiratory tract infection, can result in fever, cough, breathlessness, and wheezing. These conditions can be confused easily with bronchial foreign body, asthma, pneumonia, and other bronchial diseases, leading to an incorrect or missed diagnosis 8,25,26 . Lack of realization of the possibility of a tumor and the consequent need for chest CT and bronchoscopy can delay treatment, resulting in extensive adhesions and involvement of surrounding tissues, or even spread to distant airways, thereby worsening the prognosis 13,15 . PTT should be considered in patients with chronic cough, wheezing, or progressive dyspnea who respond poorly to conventional treatment. A thorough examination of the respiratory tract, especially chest CT and bronchoscopy, may reduce the likelihood of misdiagnosis, and improve treatment and prognosis.