Optimal Pathological Prediction of Vocal Fold Leukoplakia Based on IPCLs and Morphology

ABSTRACT

Objectives

To establish a combined classification based on intrapapillary capillary loops (IPCLs) and morphological characteristics to improve the accuracy of pathological prediction of vocal fold leukoplakia (VFL).

Materials and Methods

A prospective research was conducted of VFL patients diagnosed by IPCLs and morphology. The VFL cases were classified as Type I–III based on IPCLs and morphological characteristics. Type I referred to VFL with dendritic vessels but not IPCLs. Type II defined VFL without any IPCLs or vessels and classified by morphology into two subtypes as nonrough Type IIa and rough Type IIb. Type III referred to VFL with IPCLs and classified into two subtypes as Type IIIa with small IPCLs and Type IIIb with large IPCLs or vascular distortion in or around lesions. Predicting pathology accuracy was analyzed.

Results

One hundred and eighty‐two eligible patients were recruited. The prediction accuracy rates of VFL pathology were 81.5% according to the 2019 Ni classification. The combined classification included four cases of Type I, 28 Type IIa, 35 Type IIb, 56 Type IIIa, and 59 Type IIIb VFLs. The prediction accuracy rate of combined classification ranged from 95.1% to 97.3% in three observers. The average sensitivity, specificity, positive predictive value, and negative predictive value were 97.8%, 86.2%, 97.4%, and 88.2%, respectively. The interobserver agreement varied from 84.1% to 94.0%; the mean area under curve of receiver operating curve analysis was 0.954.

Conclusions and Significance

The new combined classification based on IPCLs and morphological characteristics could predict the pathology of VFL accurately.

Level of Evidence

4

The existing classifications of narrow band imaging (NBI) and morphology had achieved positive results in predicting pathology of vocal fold leukoplakia (VFL); they also had certain defects that limit clinical application. We tried to establish a new classification based on combined IPCLs typing in NBI and morphological characteristics in WLI, which achieved higher prediction accuracy and consistency in pathology prediction of 182 VFL patients.

1. Introduction

Vocal fold leukoplakia (VFL) is the white plaque that appears on the vocal cords and cannot be diagnosed as other disorders, whose pathology is hyperkeratotic or hypokeratotic epithelium with or without dysplasia [1]. Epidemiological data in the last decade suggest that the prevalence of VFL was about 1%–6% [2], with a recurrence rate of 16%–22.7% after surgery [3, 4] and a large‐span malignant transformation rate of 5%–36% due to differences in demographics [5]. Guidance on the standardized treatment of VFL has not been issued worldwide, mainly because there is still no effective method to predict the pathological properties of VFL before surgery. Although many pathological statistics showed that VFL without dysplasia accounts for nearly half of the total cases [6, 7], cancerphobia and inaccurate pathological assessment still lead to overtreatment of a large number of VFL patients, resulting in permanent damage to voice quality.

Endoscopy is an effective way to evaluate the extent and degree of VFL, in which white light imaging (WLI), stroboscopic laryngoscopy, and narrow band imaging (NBI) are the mainstream inspection methods. WLI is the most popular and intuitive inspection method, which can accurately describe the morphological characteristics of leukoplakia [8]. The advantage of stroboscopic laryngoscopy is that it can observe the vibration of vocal cord mucosa, so as to find hidden lesions under white light [9]. The characteristic of NBI is to highlight the development of local blood vessels; epithelial dysplasia of VFL is often accompanied by vertical vascular proliferation called intrapapillary capillary loops (IPCLs), so punctiform vessels can be observed under NBI, which has a unique advantage in clarifying the extent of lesions [10].

Many scholars had tried to predict the pathology of VFL through endoscopy, and standardized the prediction method through multilevel classification. We previously proposed three classifications based on the morphology of leukoplakia, which had high interobserver agreement for pathological grading and significant correlation with pathological outcomes [8]. Combined use of morphological classification with strobe laryngoscopes has been found to fairly estimate the degree of dysplasia in VFL [11]. The five‐type classification of laryngeal lesions under the NBI [12] and the six‐type classification specific to VFL [13] were considered to be predictors of pathology with more than 90% prediction accuracy [13, 14].

Although the existing classifications had achieved positive results in retrospective application, they also had certain defects that limit wider clinical application: elevated and smooth VFL in morphological classification and Type II in Ni six‐type NBI classification shared the problem of significant pathologic variation and inaccurate pathological prediction. Therefore, in this research, we tried to establish a new classification based on combined IPCLs typing in NBI and morphological characteristics in WLI to make up for the above defects, further improve the accuracy of pathological prediction, and simplify the classification as much as possible to facilitate memory and clinical application of endoscopists and otolaryngologists.

2. Materials and Methods

This research was a prospective study approved by the ethics committee of the Eye, Ear, Nose, and Throat Hospital of Fudan University, Shanghai, China (No. 2017042‐1). All patients included in the study expressed informed consent to the use of their clinical information and imaging data.

2.1. Clinical Information

The patient's basic information was retrieved from the inpatient medical record system. Patient recruitment criteria included: (1) clinical diagnosis of unilateral or bilateral VFL and associated pathologic diagnosis; (2) nonpregnant patients ranging from 26 to 86 years old; (3) complete preoperative information, NBI data and treatment process; (4) surgical treatment and postoperative follow‐up completed by the same treatment team. Patient exclusion criteria included: (1) other causes of vocal cord white lesions, including acute laryngitis, laryngeal fungal infection caused by long‐term inhalation of corticosteroids, laryngeal tuberculosis, and other nonkeratinized vocal fold white lesions. These cases had a clear history of colds or asthma or chronic obstructive pulmonary disease or tuberculosis. The white lesions of pseudomembrane and fungal infection were more delicate and not as rough as leukoplakia. The white lesions of laryngeal tuberculosis were diffuse, granulomatous, and accompanied by a lot of secretions; (2) patients who were out of age range or unable to undergo surgical treatment due to other reasons; (3) patients with missing preoperative information, laryngoscopy data or postoperative pathology.

2.2. WLI and NBI Examination

WLI was performed in the outpatient laryngoscopy room using flexible endoscopy; the specific model was ENF TYPE VT2 and TYPE VH (Olympus Medical Systems Corp, Tokyo, Japan). Then NBI mode was switched to observe the IPCLs. The specific model was CV‐190 (Olympus Medical Systems Corp, Tokyo, Japan). After the endoscopists completed the relevant examination, they would write both electronic and paper versions of the examination reports; the otolaryngologists could choose the form of obtaining the examination reports.

The morphological classification [8] and Ni classification [13] were employed as previous reports. The morphological classification included flat and smooth, elevated and smooth, rough types, which were based on the texture of leukoplakia and whether they protruded from the surface of the vocal cords. The Ni classification contained Type I–VI, Type I with obliquely running vessels but not IPCLs, Type II without vessels or IPCLs, Type III with small IPCLs, Type IV with large IPCLs on the surface of white plaque, Type V with large IPCLs at the surface of the vocal cord mucosa outside the leukoplakia, and Type VI with large IPCLs or twisted earthworm‐like vessels.

The pathologies of flat and smooth, elevated and smooth types in morphological classification and Type I, II, III in Ni classification were precursor lesions, which contained non‐, mild, and moderate dysplasia. The pathologies of rough type in morphological classification and Type IV, V, VI in Ni classification were advanced precancerous lesions (APL) or carcinoma, which contained severe dysplasia, carcinoma in situ, and carcinoma.

2.3. New Combined Classification Based on Morphology and IPCLs

The new combined classification was based on IPCLs and morphological characteristics, which was divided into the following three types, of which Types II and III were each divided into two subtypes as shown in Figure 1:

• Type I: VFL with dendritic vessels but without IPCLs (Figure 1A,B).

• Types II: VFL with absence of vessels and IPCLs, subtyped by morphological characteristics.

  • ■
    IIa: Smooth lesions with flat (Figure 1C,D) or elevated (Figure 1E,F) surface. Flat and smooth type: Surface: smooth; Margin: lesion without raised margins, being continuous with the surrounding mucosa; Texture: homogeneous, regular, the lesion with even coloration. Elevated and smooth type: Surface: smooth; Margin: lesion with raised margins, sharply demarcated from the surrounding mucosa; Texture: homogeneous, regular, the lesion with even coloration.
  • ■
    IIb: Rough (Figure 1G,H). Surface: wrinkled, corrugated; Margin: lesion with raised margins, sharply demarcated from the surrounding mucosa; Texture: nonhomogeneous, irregular, the lesion with uneven coloration and is usually accompanied with erosion or ulceration.
• Types III: VFL with IPCLs.

  • ■
    IIIa: Small, scattered, evenly arranged (Figure 1I,J).
  • ■
    IIIb: Coarse, dense, uneven arrangement (Figure 1K,L) or vascular distortion, dilation (Figure 1M,N).

FIGURE 1.

Combined classification of vocal fold leukoplakia based on IPCLs and morphology. Type I with dendritic vessels but without IPCLs in WLI (A) and NBI (B). Types II with absence of vessels and IPCLs, subtyped by morphology as Types IIa with smooth and flat lesion in WLI (C) and NBI (D) or smooth and elevated lesion in WLI (E) and NBI (F), Types IIb with rough lesion in WLI (G) and NBI (H). Types III with existence of IPCLs, N‐types IIIa with small, scattered, evenly arranged IPCLs in WLI (I) and NBI (J), Types IIIb with coarse, dense, uneven arrangement IPCLs in WLI (K) and NBI (L) or vascular distortion, dilation in WLI (M) and NBI (N). WLI, white light imaging. NBI, narrow‐band imaging.[Color figure can be viewed in the online issue, which is available at www.laryngoscope.com]

Dendritic vessels referred to the vessels that run parallel to the surface of the vocal cords and branched out in a dendritic manner. Vascular distortion referred to the thickening of the vessels on the surface of the vocal cords, which were segmented and earthworm‐like twisted.

In new combined classification, the pathology of Type I, IIa VFLs was hypothesized to be precursor lesions, containing hyperkeratosis or parakeratosis with non‐, mild, moderate dysplasia; Type IIIa was mainly hypothesized to be APL (severe dysplasia/carcinoma in situ), and IIb, IIIb were hypothesized to be APL or carcinoma.

2.4. Treatment and Follow‐Up

The treatment and follow up of the VFL was seen as a flowchart in Figure 2. The Type I VFLs were mainly treated conservatively. Surgical treatments were recommended for patients with Type IIb, IIIa, IIIb lesions and VFLs with ineffective conservative treatments. Patients with Type IIa were provided with the option of the conservative or surgical treatment strategy.

FIGURE 2.

Treatment and follow‐up flowchart of vocal fold leukoplakia according to combined classification. #, voice rest, cessation of smoking and alcohol consumption, and medications such as proton pump inhibitors to treat reflux disease. *, transoral CO₂ laser‐assisted microlaryngoscopy.

Conservative treatments included voice rest, quitting smoking and alcohol consumption, and medications such as proton pump inhibitors to treat reflux disease. Patients were advised to have monthly follow‐up visits and laryngoscopy in the first 3 months after diagnosis. If the disease resolved, the patients were advised to maintain healthy lifestyles and return for follow‐up visits every 3–6 months for the next 3 years. If the lesions remained at Type I stage, office‐based potassium–titanyl–phosphate (KTP) laser was suggested. If disease progressed to Type II or III, the patients received treatments and follow‐up as described below. The latter two situations were defined as cases with ineffective conservative treatments.

All patients' outpatient and surgical treatments were completed by the same treatment team. The surgery was performed under conventional general anesthesia and a self‐retaining laryngoscope. A binocular microscope (ZEISS S88, Carl Zeiss Shanghai Co.) was employed to fully expose the glottis area; the resection of epithelium was performed with carbon dioxide laser (SmartXide², DEKA, Italy) and a resection range > 1 mm.

All patients were discharged from hospital on the second day after surgery and were advised to return to the outpatient clinic for follow‐up and surveillance of laryngoscopy in the first 6 months after surgery. If the operative area epithelialized well and no recurrence was observed, the patients were advised to maintain healthy lifestyles and return for follow‐up every 3–6 months for the next 3 years. If Type I, IIa VFLs were observed in the follow‐up period, office‐based KTP laser was suggested in well‐equipped medical units; otherwise, surgery was recommended as soon as possible. Recurrent Type IIb, IIIa, IIIb lesions were suggested for transoral microlaryngoscopy.

2.5. Pathology and Correlation With Imaging Classification

Postoperative pathology was classified by senior doctors from the pathology department according to the 2005 WHO standard as hyperkeratosis or parakeratosis with non‐, mild, moderate, severe dysplasia/carcinoma in situ, carcinoma [15]. Referring to the method in 2019 Ni classification [13], VFLs with non‐, mild, moderate dysplasia were defined as precursor lesions; others with severe dysplasia/carcinoma in situ, carcinoma were defined as APC+ carcinoma.

Two otolaryngological surgeons (O1, O2) who specialized in VFL research and one endoscopist (E) who specialized in laryngoscopy with more than 5‐year experience participated in the research. They performed pathological prediction on WLI and NBI images of VFL according to the above classification. The predicting pathology accuracy rates, sensitivity, specificity, positive predictive value, negative predictive value, interobserver agreement, area under curve (AUC) of receiver operating curve (ROC) were analyzed.

2.6. Statistical Analysis

The statistical calculations of this research were completed by SPSS version 22.0 software (IBM Corporation, Chicago, IL, USA). Except that the comparison of age between groups was completed by one‐way ANOVA, the rest of the statistical comparisons including the baseline data, NBI classification correspondence with pathology, were completed by R*C chi squared test. The interobserver agreement was calculated by Cohen's κ. The predictive value evaluation of new classification was completed by ROC analysis and AUC was calculated. Statistically significant was defined with two‐sided p < 0.05.

3. Results

A total of 182 eligible patients ranging from 45 to 70 years old were involved in this research, including four patients with pathological diagnosis of nondysplasia, 12 patients with mild dysplasia, 16 patients with moderate dysplasia, 34 patients with severe dysplasia/carcinoma in situ, and 116 patients with carcinoma. Grouped according to the pathology, there was no statistical difference in baseline information, including age, gender (p = 0934, 0.762, respectively).

The distribution of pathology within the three groups of morphological classification was shown in Table 1. Flat and smooth VFLs contained 3 cases with nondysplasia, 2 cases with mild dysplasia, which all belonged to precursor lesions. Rough VFLs showed a concentrated distribution trend of APL, 18 cases were severe dysplasia/carcinoma in situ, 111 cases were carcinoma, other 2 cases were moderate dysplasia. Elevated and smooth VFLs showed relatively scattered pathologies, including 1 case with nondysplasia, 10 cases with mild dysplasia, 14 cases with moderate dysplasia, 16 cases with severe dysplasia/carcinoma in situ, 5 cases with carcinoma. According to the morphological classification, the average prediction accuracy rate of morphology was 89.0% (87.4%–90.7%). The average sensitivity, specificity, positive predictive value, negative predictive value was 88.3% (86.3%–90.3%), 92.9% (92.6%–93.1%), 98.5% (98.5%–98.6%), 59.4% (56.3%–62.5%), respectively (Figure 3).

TABLE 1.

Pathological distribution of vocal fold leukoplakia in morphological classification.

	Precursor lesion			APL + carcinoma		p
	Nondysplasia	Mild dysplasia	Moderate dysplasia	Severe dysplasia/carcinoma in situ	Carcinoma
Morphological classification						< 0.001
Flat and smooth	3	2	0	0	0
Elevated and smooth	1	10	14	16	5
Rough	0	0	2	18	111

Abbreviation: APL, advanced precancerous lesion.

FIGURE 3.

Summary of statistical indicators of the morphological, Ni and new classification. PPV, positive predictive value. NPV, negative predictive value.

According to 2019 Ni classification, as listed in Table 2, the case number of Type I–VI was 3, 64, 2, 82, 18, 13, respectively. The pathological distribution of Type I, III, IV–VI was closely consistent with the predicted pathological types, as the former two were precursor lesions and Type IV–VI were APLs and carcinoma. The Type II VFLs contained 1 case with nondysplasia, 12 with mild dysplasia, 14 with moderate dysplasia, 9 APL, and 28 carcinomas. The average prediction accuracy rate of Ni classification was 81.5% (79.7%–84.1%). The average sensitivity, specificity, positive predictive value, negative predictive value was 77.8% (75.3%–80.7%), 99.0% (97.3%–100.0%), 99.7% (99.1%–100.0%), and 48.7% (46.4%–52.5%), respectively (Figure 3).

TABLE 2.

Pathological distribution of vocal fold leukoplakia in 2019 Ni classification.

	Precursor lesion			APL + carcinoma		p
	Nondysplasia	Mild‐dysplasia	Moderate‐dysplasia	Severe‐dysplasia/carcinoma in situ	Carcinoma
2019 Ni classification						< 0.001
Type I	3	0	0	0	0
Type II	1	12	14	9	28
Type III	0	0	2	0	0
Type IV	0	0	0	23	59
Type V	0	0	0	1	17
Type VI	0	0	0	1	12

Abbreviation: APL, advanced precancerous lesion.

As listed in Table 3, reanalysis of all cases using the combined classification revealed that Type I contained 3 cases with nondysplasia and 1 case with mild dysplasia. In subgroups of Type II, the pathologies of Type IIa included 1 case of non‐dysplasia, 10 cases of mild dysplasia, 14 cases of moderate dysplasia, and 3 cases with APL. While Type IIb VFLs were mainly APL + carcinoma, which contained 2 moderate‐dysplasia cases, 6 cases with APL, and 27 cases with carcinoma. Subgroups of Type III were mainly APL + carcinoma except Type IIIa contained 1 mild‐dysplasia case. The numbers of cases with APL/carcinoma in Type IIIa and IIIb were 18/37 and 7/52, respectively.

TABLE 3.

Pathological distribution of vocal fold leukoplakia in combined classification.

	Precursor lesion			APL + carcinoma		p
	Nondysplasia	Mild dysplasia	Moderate dysplasia	Severe dysplasia/carcinoma in situ	Carcinoma
Combined classification						< 0.001
Type I	3	1	0	0	0
Type II
IIa	1	10	14	3	0
IIb	0	0	2	6	27
Type III
IIIa	0	1	0	18	37
IIIb	0	0	0	7	52

Abbreviation: APL, advanced precancerous lesion.

The prediction accuracy rate of the combined classification was 96.0%, which ranged from 95.1% to 97.3% in three observers. The average sensitivity, specificity, positive predictive value, negative predictive value was 97.8% (97.4%–98.0%), 86.2% (79.3%–93.1%), 97.4% (96.2%–98.7%), 88.2% (86.2%–90.0%), respectively (Figure 3). The Cohen's κ values and interobserver agreement value of morphological classification (0.91, 0.79, 0.88; 96.7%, 92.3%, 95.6%), Ni classification (0.77, 0.63, 0.80; 72.0%, 83.0%, 85.2%), new combined classification (0.62, 0.65, 0.76; 84.1%, 90.1%, 94.0%) was calculated in E vs. O1, E vs. O2, O1 vs. O2, respectively (Table 4). As shown in Figure 4, ROC curve revealed the mean AUC of morphological classification was 0.902, the counterpart of Ni classification was 0.865, the data of combined classification used by three independent observers were 0.950, 0.935, 0.977, respectively.

TABLE 4.

Interobserver agreement between the reviewers.

Observer pair	Cohen's κ			Agreement (%)
	Morphological classification	Ni classification	New combined classification	Morphological classification	Ni classification	New combined classification
E and O1	0.91	0.77	0.62	96.7	72.0	84.1
E and O2	0.79	0.63	0.65	92.3	83.0	90.1
O1 and O2	0.88	0.80	0.76	95.6	85.2	94.0

Abbreviations: E, endoscopist; O, otolaryngologist.

FIGURE 4.

Receiver operating curve analysis of morphological, Ni and new classification. Red line, morphological classification. Green line, Ni classification. Purple and orange line, two otolaryngologist observers. Blue line, one endoscopist observer.[Color figure can be viewed in the online issue, which is available at www.laryngoscope.com]

4. Discussion

The purpose of this research was to construct a combined classification criteria by integrating existing classification methods to improve the accuracy of preoperative laryngoscopy in predicting the pathology of VFL. The existing classifications were mainly based on the variable WLI and NBI performances with different pathological degrees. In this research, the prediction accuracy rate of the new combined classification was 96.0%, morphological classification under WLI achieved 89.0%, while the counterpart of 2019 Ni classification under NBI was 81.5%. The reason of highest accuracy achieved by the new combined classification was due to the employment of the main functional characteristics of morphological and NBI classification. The reason for the lower accuracy of the latter was mainly due to the wider pathological distribution of Type II lesions; in original research, the predicted pathology of Type II was “benign leukoplakia.” While in our research, according to the 2019 Ni classification, Type II cases contained 29 cases in the precursor part and 37 cases in the APL + carcinoma part. The combined classification achieved higher sensitivity and negative predictive values than the previous two classifications. The advantage of the new classification in terms of sensitivity mainly came from the simultaneous application of rough morphology and IPCL which clearly indicated APL and carcinoma. On the contrary, leukoplakia with smooth morphology and no IPCL were probably precursor lesions, which also guaranteed a higher negative predictive value.

The attempt of this research was to combine the prediction characteristics of IPCLs and morphology in order to compensate for each other and overcome the existing drawbacks. In the comparison of prediction results of different classifications, 23 cases were incorrectly identified with the morphological classification, 34 cases were incorrectly identified with the Ni classification, but correctly identified with the new classification, which were all concentrated in Type IIa of the new classification, indicating the advantage of morphological classification in the absence of IPCL. The misdiagnosed cases in the morphological classification were not the same as Type IIa cases in the new classification. Most of the former were in Type III, where IPCLs could be observed and diagnosed as APL or carcinoma based on the distribution characteristics of IPCLs.3 elevated and smooth cases with severe dysplasia were misjudged, indicating that the pathology of leukoplakia without IPCLs and morphological tendency was still difficult to predict. There was no case being incorrectly identified with the new classification but predicted by the morphological or Ni classification.

The combined classification showed 84.1%–94.0% of interobserver agreement values, mean 0.954 of AUC under the ROC analysis by three independent observers, which suggested it may be a reliable prediction system. There was a higher percent agreement between the two‐otolaryngologist pairing. The endoscopist observed the laryngoscopic performance of VFL dynamically during the examination, while the otolaryngologist could only make an assessment based on the laryngoscopic reports. There might be certain differences in their judgments, which is the reason that endoscopists should be included in this research. Nevertheless, there was still some inconsistency among the three observers in the Type II lesions. The main reason was that the Type II classification standard was no IPCL; however, not every patient could cooperate well during the examination to allow the laryngoscope to get close enough to observe the IPCL around the lesions.

In this research protocol, surgery under general anesthesia was recommended for patients suspected of APL and carcinoma based on laryngoscopic classification, rather than after outpatient pathological diagnosis by biopsy under local anesthesia. The main reasons were in concern of the shortcomings of inaccurate biopsy location, insufficient biopsy depth, and diagnostic bias caused by different degrees of pathology in different subregions, which might misdiagnose and delay surgical treatment for VFL patients. We classified pathological prediction objects into precursor lesions (non‐, mild, moderate dysplasia) and APL (severe dysplasia, carcinoma in situ) or carcinoma for the following reasons. The prognosis of non‐, mild, and moderate dysplasia was similar; conservative treatment might be effective, and the lesions' progression was reversible, while severe dysplasia, carcinoma in situ, and carcinoma required surgery as soon as possible. Therefore, the two‐tier pathological prediction method was easy for clinicians to remember and apply.

In the early stages of designing the new classification, whether VFL with small IPCLs with relatively regular arrangement and without clear boundaries should be considered APL had been carefully considered. The essence of IPCLs was the obliquely extending blood vessels emanating from the mucosal vascular network above the lamina muscularis mucosae [16]. The emergence and development of IPCLs were closely related to local inflammation and pro‐angiogenic growth factors produced by progressive lesions [17]. Although laryngeal inflammation could also lead to IPCLs [18], the discovery of IPCLs around precancerous lesions such as leukoplakia could indicate the possibility of malignancy. The umbrella effect of the white plaque blocked the large and distorted IPCLs, and only small IPCLs around the lesion could be observed, which might lead to the illusion of only small IPCLs. Also, Rzepakowska et al. reviewed 62 VFL patients and suggested the strong link between atypical vascular patterns in the surrounding mucosa and progressive dysplasia [19]. Furthermore, there was a certain correlation between the determination of small or large IPCLs and the distance between endoscopy and lesions during laryngoscopy. Therefore, there were subjective factors in determining the size of IPCLs. In our research, the case numbers of Type III in 2019 Ni classification were significantly smaller than Ni original report [13]. It was partly due to the influence of lesion distribution and treatment process, and the clarity of fiberoptic laryngoscopy might also relate to this phenomenon.

In addition to the traditional prediction of VFL pathology by otorhinolaryngologists and laryngoscopists based on morphology or NBI classifications, new methods based on deep‐learning algorithms also received achievement in recent years. You et al. employed six classical deep learning models to retrospectively perform WLI and NBI classification, in which GoogLeNet, DenseNet‐121, and ResNet‐152 performed excellently and acquired highest overall accuracy of WLI to be 0.9583, NBI with 0.9478 [20]. The theoretical basis of artificial intelligence intervention for pathology prediction was based on existing algorithms, where laryngoscopic images of lesions with correspondence pathologies were uploaded to train the algorithm. The trained algorithm was then blindly tested on another set of laryngoscopic images with known pathologies to assess its prediction accuracy [21]. It could therefore be argued that the advantages of deep learning algorithms were the more efficient training methods and shorter training times. However, the current learning logic of artificial intelligence could be applied with existing typing prediction methods, which was unable to design a more practical classification. There is still necessity of briefer and more effective methodology for VFL pathological prediction.

There were several shortcomings in this research. Firstly, the proportions of the pathological distribution of the patients included differed from the epidemiological distribution of VFL [22], in which non‐ and mild dysplasia accounted for more than 50% of pathologies. Most of the cases in this research being APL and carcinoma might result in potential selection bias. The reason was the more stringent inclusion criteria for surgery at our center, where potentially non‐, mild, and moderate dysplasia cases that resolved significantly with conservative and pharmacological treatments were not treated with surgery. Further multicenter prospective studies will address this issue and re‐validate the value of the new classification. Secondly, the age range of patients included in the study was narrower than in previous studies for the following reasons: (1) due to the popularization of VFL scientific knowledge and the decrease in the number of young smokers, there were fewer patients aged 26–44 years old who need surgical treatment. The vocal fold white lesions found in young people were mostly caused by inflammatory exudate in acute laryngitis. The baseline analysis of this population was quite different, which would affect the statistics, so it was not included in the study; (2) there were indeed patients with VFL aged 71–86 years, but due to the special nature of our hospital as a specialist hospital, we lacked auxiliary support from the department of internal medicine and had limited ability to perform perioperative management of elderly people with multiple medical diseases, so there were relatively few patients in this age group. Thirdly, the participants in the research of the application of the new classification were all senior otorhinolaryngologists and laryngoscopists with long‐term research of VFL. The mastery of the application among doctors at different stages of their careers and the corresponding learning curves remain to be studied.

5. Conclusion

The new combined classification based on IPCLs and morphological characteristics could predict pathology accurately.

Funding

This research is supported by the Science and Technology Commission of Shanghai Municipality of China (Grant numbers 21Y11912000 and 22ZR1409800) and also supported by the National Natural Science Foundation of China (Grant numbers 82271149 and 82102863).

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.