Learning curve for endoscopic evaluation of vocal folds lesions with narrow band imaging

Improving real-time detection of laryngeal lesions in endoscopic images using a decoupled super-resolution enhanced YOLO

BACKGROUND AND OBJECTIVE

Laryngeal Cancer (LC) constitutes approximately one third of head and neck cancers. Detecting early-stage lesions in this anatomical region is crucial for achieving a high survival rate. However, it poses significant diagnostic challenges owing to the varied appearance of lesions and the need for precise characterization for appropriate clinical management. Conventional diagnostic approaches rely heavily on endoscopic examination, which often requires expert interpretation and may be limited by subjective assessment. Deep learning (DL) approaches offer promising opportunities for automating lesion detection, but their efficacy in handling multi-modal imaging data and accurately localizing small lesions remains a subject of investigation. Furthermore, the clinical domain may largely benefit from the deployment of efficient DL methods that can ensure equitable access to advanced technologies, regardless of the availability of resources that can often be limited. In this study, a DL-based approach, named SRE-YOLO, was introduced to provide real-time assistance to less-experienced personnel during laryngeal assessment, by automatically detecting lesions at different scales from endoscopic White Light (WL) and Narrow-Band Imaging (NBI) images.

METHODS

During the training, the SRE-YOLO integrates a YOLOv8 nano (YOLOv8n) baseline with a Super-Resolution (SR) branch to enhance lesion detection. This last component is decoupled during inference to preserve the low computational demand of the YOLOv8n baseline. The evaluation was conducted on a multi-center dataset, encompassing diverse laryngeal pathologies and acquisition modalities.

RESULTS

The SRE-YOLO method improved the Average Precision (AP@IoU=0.5) in lesion detection by 5% with respect to the YOLOv8n baseline, while maintaining the inference speed of 58.8 Frames Per Second (FPS). Comparative analyses against state-of-the-art DL methods highlighted the efficacy of the SRE-YOLO approach in balancing detection accuracy, computational efficiency, and real-time applicability.

CONCLUSIONS

This research underscores the potential of SRE-YOLO in developing efficient DL-driven decision support systems for real-time detection of laryngeal lesions at different scales from both WL and NBI endoscopic data.

Diagnostic Assessment (Imaging) and Staging of Laryngeal Cancer

Diagnosis of larynx cancer relies on a detailed history and physical and objective assessment with endoscopy and imaging. Endoscopy is needed to assess for vocal fold function that directly affects staging. Computed tomography and MRI can be used to assess for tumor extent in relation to intra- and extra-laryngeal structures, especially paraglottic and pre-epiglottic space involvement as well as cartilage invasion. Accurate staging is critical for subsequent treatment decision-making regarding larynx preservation.

From Hype To Reality: A Narrative Review on the Promising Role of Artificial Intelligence in Larynx Cancer Detection and Transoral Microsurgery

Early larynx cancer detection plays a crucial role in improving treatment outcomes and recent studies have shown promising results in using artificial intelligence for larynx cancer detection. Artificial intelligence also has the potential to enhance transoral larynx microsurgery. This narrative review summarizes the current evidence regarding its use in larynx cancer detection and potential applications in transoral larynx microsurgery. The utilization of artificial intelligence in larynx cancer detection with white light endoscopy and narrow-band imaging helps improve diagnostic accuracy and efficiency. It can also potentially enhance transoral larynx microsurgery by aiding surgeons in real-time decision-making and minimizing the risk of complications. However, further prospective studies are warranted to validate the findings, and additional research is necessary to optimize the integration of artificial intelligence in our clinical practice.

Role of Narrow Band Imaging in Laryngeal Lesions: A Prospective Study from Southern India

To assess the utility of Narrow Band Imaging (NBI) as a diagnostic tool in evaluating laryngeal pathology in patients presenting with hoarseness. Study design: Prospective cross sectional diagnostic study. Methods: 200 patients with voice change were recruited and laryngeal findings documented with high definition flexible endoscopy with both white light and narrow band light and the representative still images recorded for analysis. The NBI intraepithelial papillary capillary loop (IPCL) patterns was compared with the histopathology report. Of the 200 patients evaluated, 84 lesions were biopsied which included both benign and malignant lesions. The sensitivity obtained was 73.3% [54.1-87.7% with 95% CI] and the specificity was 87% [75.1-94.6% with 95% CI] for detecting malignant lesions. NBI can be considered as a useful diagnostic tool in evaluating laryngeal pathology and can be used to detect early premalignant and malignant lesions.

Videolaryngoendoscopic and Stroboscopic Evaluation in Predicting the Malignancy Risk of Vocal Fold Leukoplakia

Background: Vocal fold leukoplakia (VFL), despite our knowledge of its etiopathogenetic factors, and the development of laryngeal visualization, remains a diagnostic and therapeutic challenge. Objective: This research aimed to explore the efficacy of clinical and morphological feature identification in videolaryngoendoscopy (VLE) using a three-tier classification, and videolaryngostroboscopy (VLS) in predicting the risk of VFL malignant transformation. Material and Methods: We examined 98 patients with VFL by flexible endoscopy under VLE and VLS. Morphological characteristics of 123 lesions including the surface, margin, and texture were assessed; then, VFL was subdivided into three types: I—flat and smooth, II—elevated and smooth, and III—rough. Based on the histopathological findings, 76 (61.79%) lesions were classified as low- and 47 (38.21%) lesions as high-grade dysplasia. Results: The inter-rater agreement between two raters evaluating the VFL in VLE was almost perfect (Cohen’s kappa = 0.826; p < 0.00; 95%CI 0.748−0.904). In ROC curve analysis, the AUC difference between Rater I and Rater II was 0.024 (0.726 vs. 0.702). In multivariate analysis, high-risk VFL was positively related to unilateral plaque localization (p = 0.003), the type III VLE classification (p = 0.013), absence of a mucosal wave (p = 0.034), and a positive history of alcohol consumption (p = 0.047). In ROC analysis, VLE had an AUC of 0.726, with a high sensitivity of 95.7% and low specificity of 40.8%. The NPV was high, at 93.9%; however, the PPV was low, at 50%. The proposed logistic regression model including features significant in multivariate analysis showed lower sensitivity (80.9% vs. 95.7%) and lower NPV (86.2% vs. 93.9%); however, the specificity and PPV were improved (73.7% vs. 40.8% and 65.5% vs. 50.0%, respectively). Conclusions: The combination of clinical history with endoscopic (plaque morphology) and stroboscopic examination (mucosal wave assessment) can fairly estimate the degree of dysplasia in VFL and thus is recommended for use in clinical settings. The findings of this study can be used to guide the decision regarding immediate biopsy or watchful waiting.

Artificial Intelligence in Laryngeal Endoscopy: Systematic Review and Meta-Analysis

Background: Early diagnosis of laryngeal lesions is necessary to begin treatment of patients as soon as possible to preserve optimal organ functions. Imaging examinations are often aided by artificial intelligence (AI) to improve quality and facilitate appropriate diagnosis. The aim of this study is to investigate diagnostic utility of AI in laryngeal endoscopy. Methods: Five databases were searched for studies implementing artificial intelligence (AI) enhanced models assessing images of laryngeal lesions taken during laryngeal endoscopy. Outcomes were analyzed in terms of accuracy, sensitivity, and specificity. Results: All 11 studies included presented an overall low risk of bias. The overall accuracy of AI models was very high (from 0.806 to 0.997). The accuracy was significantly higher in studies using a larger database. The pooled sensitivity and specificity for identification of healthy laryngeal tissue were 0.91 and 0.97, respectively. The same values for differentiation between benign and malignant lesions were 0.91 and 0.94, respectively. The comparison of the effectiveness of AI models assessing narrow band imaging and white light endoscopy images revealed no statistically significant differences (p = 0.409 and 0.914). Conclusion: In assessing images of laryngeal lesions, AI demonstrates extraordinarily high accuracy, sensitivity, and specificity.

The Narrow Band Imaging as an essential complement to White Light Endoscopy in Recurrent Respiratory Papillomatosis diagnostics and follow-up process

The aim of the study is to discuss the role of NBI imaging in upper respiratory papillomatosis (RRP). In the available literature on the subject, as of 2009, there have been 117 publications on NBI imaging focusing on the diagnosis of the upper respiratory and digestive tract, especially the larynx. They discuss the following diagnostic and therapeutic problems: identifying an abnormal vascular pattern within the mucosa, differentiating benign and malignant lesions, assessing the nature of lesions under the leukoplakia plaque, pre-operative definition of laryngeal cancer margins, supporting laser procedures to obtain safe margins, postoperative monitoring after cancer treatment, detecting second simultaneous and metachronic neoplasms and primary tumors in the case of metastases from an unknown primary site. Few reports have been devoted to the assessment of the extent and recurrence of RRP. Due to its recurrent nature, this pathology deserves special attention as it requires multiple inspections and manipulation within the larynx. It also applies to the pediatric population, where there should be no margin for underestimating or overestimating changes, because any suspicion of recurrence results in subsequent general anesthesia. Hence the attempt to define an unambiguous picture of RRP recurrence and the criteria for referring for subsequent surgical treatment.

The Impact of Narrow-band Imaging on the Pre- and Intra- operative Assessments of Neoplastic and Preneoplastic Laryngeal Lesions. A Systematic Review

Introduction  Narrow-band imaging is an endoscopic diagnostic tool that, focusing on superficial vascular changes, is useful to detect suspicious laryngeal lesions, enabling their complete excision with safe and tailored resection margins.

Objectives  To analyze the applications and benefits of narrow-band imaging in detecting premalignant and malignant laryngeal lesions through a comparison with white-light endoscopy.

Data Synthesis  A literature search was performed in the PubMed, Scopus and Web of Science databases using strict keywords. Then, two authors independently analyzed the articles, read the titles and abstracts, and read completely only the relevant studies according to certain eligibility criteria. In total, 14 articles have been included in the present review; the sensitivity, specificity, positive and negative predictive values, and accuracy of pre- and/or intraoperative narrow-band imaging were analyzed. The analysis showed that narrow-band imaging is better than white-light endoscopy in terms of sensitivity, specificity, positive and negative predictive values, and accuracy regarding the ability to identify cancer and/or precancerous laryngeal lesions. Moreover, the intraoperative performance of narrow-band imaging resulted more effective than the in-office performance.

Conclusion  Narrow-band imaging is an effective diagnostic tool to detect premalignant and malignant laryngeal lesions and to define proper resection margins. Moreover, narrow-band imaging is useful in cases of leukoplakia that may cover a possible malignant lesion and that cannot be easily assessed with white-light endoscopy. Finally, a shared, simple and practical classification of laryngeal lesions, such as that of the European Laryngological Society, is required to identify a shared lesion management strategy.

Key Points

Effect of a training course on the diagnosis of vocal fold leukoplakia by narrow-band imaging

OBJECTIVE

To investigate the value of narrow-band imaging training for differentiating between benign and malignant vocal fold leukoplakia.

METHOD

Thirty cases of vocal fold leukoplakia were selected.

RESULTS

Narrow-band imaging endoscopy training had a significant positive effect on the specificity of the differential diagnosis of vocal fold leukoplakia. In addition, the consistency of diagnostic typing of vocal fold leukoplakia by narrow-band imaging improved to 'moderate agreement' following the combination of types I and II and the combination of types IV, V and VI in the typing of vocal fold leukoplakia.

CONCLUSION

The narrow-band imaging training course may improve the ability of laryngologists to diagnose vocal fold leukoplakia. The new endoscopic diagnostic classification by narrow-band imaging needs to be further simplified to facilitate clinical application.

Use of narrowband imaging for the diagnosis and screening of laryngeal cancer: A systematic review and meta-analysis

We evaluated the diagnostic accuracy of narrowband imaging (NBI) in terms of detecting laryngeal cancer compared to that of white light endoscopy (WLE). Two reviewers individually searched the six databases for studies published between the first record date and December 31, 2019. We recorded the numbers of true positives, true negatives, false positives, and false negatives. Quality Assessment of Diagnostic Accuracy Studies ver. 2 software was used to assess the studies. The extent of the inter-rater agreement was also measured. The diagnostic odds ratio (OR) associated with NBI was 87.463 (95% confidence interval [CI]: 46.968, 160.873). The area under the summary receiver operating characteristic curve was 0.954. NBI was more diagnostically accurate than WLE, which was associated with a diagnostic OR of 13.750. NBI affords high diagnostic accuracy, thus supporting a role for NBI in the diagnostic work-up of laryngeal cancer.

Development of Learning Curves for Bronchoscopy: Results of a Multicenter Study of Pulmonary Trainees

BACKGROUND

There are currently no reference standards for the development of competence in bronchoscopy.

RESEARCH QUESTION

The aims of this study were to (1) develop learning curves for bronchoscopy skill development and (2) estimate the number of bronchoscopies required to achieve competence.

STUDY DESIGN AND METHODS

Trainees from seven North American academic centers were enrolled at the beginning of their pulmonology training. Performance during clinical bronchoscopies was assessed by supervising physicians using the Ontario Bronchoscopy Assessment Tool (OBAT). Group-level learning curves were modeled using a quantile regression growth model, where the dependent variable was the mean OBAT score and the independent variable was the number of bronchoscopies performed at the time the OBAT was completed.

RESULTS

A total of 591 OBAT assessments were collected from 31 trainees. The estimated regression quantiles illustrate significantly different learning curves based on trainees' performance percentiles. When competence was defined as the mean OBAT score for all bronchoscopies rated as being completed without need for supervision, the mean OBAT score associated with competence was 4.54 (95% CI, 4.47-4.58). Using this metric, the number of bronchoscopies required to achieve this score varied from seven to 10 for the 90th percentile of trainees and from 109 to 126 for the lowest 10th percentile of trainees. When competence was defined as the mean OBAT score for the first independent bronchoscopy, the mean was 4.40 (95% CI, 4.20-4.60). On the basis of this metric, the number of bronchoscopies required varied from one to 11 for the 90th percentile of trainees and from 83 to 129 for the lowest 10th percentile of trainees.

INTERPRETATION

We were able to generate learning curves for bronchoscopy across a range of trainees and centers. Furthermore, we established the average number of bronchoscopies required for the attainment of competence. This information can be used for purposes of curriculum planning and allows a trainee's progress to be compared with an established norm.