Instance segmentation of upper aerodigestive tract cancer: site-specific outcomes.

IF 2.1 4区医学 Q2 OTORHINOLARYNGOLOGY

Acta Otorhinolaryngologica Italica Pub Date : 2023-08-01 DOI:10.14639/0392-100X-N2336

Alberto Paderno, Francesca Pia Villani, Milena Fior, Giulia Berretti, Francesca Gennarini, Gabriele Zigliani, Emanuela Ulaj, Claudia Montenegro, Alessandra Sordi, Claudio Sampieri, Giorgio Peretti, Sara Moccia, Cesare Piazza

{"title":"Instance segmentation of upper aerodigestive tract cancer: site-specific outcomes.","authors":"Alberto Paderno, Francesca Pia Villani, Milena Fior, Giulia Berretti, Francesca Gennarini, Gabriele Zigliani, Emanuela Ulaj, Claudia Montenegro, Alessandra Sordi, Claudio Sampieri, Giorgio Peretti, Sara Moccia, Cesare Piazza","doi":"10.14639/0392-100X-N2336","DOIUrl":null,"url":null,"abstract":"Objective: To achieve instance segmentation of upper aerodigestive tract (UADT) neoplasms using a deep learning (DL) algorithm, and to identify differences in its diagnostic performance in three different sites: larynx/hypopharynx, oral cavity and oropharynx.Methods: A total of 1034 endoscopic images from 323 patients were examined under narrow band imaging (NBI). The Mask R-CNN algorithm was used for the analysis. The dataset split was: 935 training, 48 validation and 51 testing images. Dice Similarity Coefficient (Dsc) was the main outcome measure.Results: Instance segmentation was effective in 76.5% of images. The mean Dsc was 0.90 ± 0.05. The algorithm correctly predicted 77.8%, 86.7% and 55.5% of lesions in the larynx/hypopharynx, oral cavity, and oropharynx, respectively. The mean Dsc was 0.90 ± 0.05 for the larynx/hypopharynx, 0.60 ± 0.26 for the oral cavity, and 0.81 ± 0.30 for the oropharynx. The analysis showed inferior diagnostic results in the oral cavity compared with the larynx/hypopharynx (p < 0.001).Conclusions: The study confirms the feasibility of instance segmentation of UADT using DL algorithms and shows inferior diagnostic results in the oral cavity compared with other anatomic areas.","PeriodicalId":6890,"journal":{"name":"Acta Otorhinolaryngologica Italica","volume":"43 4","pages":"283-290"},"PeriodicalIF":2.1000,"publicationDate":"2023-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ftp.ncbi.nlm.nih.gov/pub/pmc/oa_pdf/be/49/aoi-2023-04-283.PMC10366566.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Acta Otorhinolaryngologica Italica","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.14639/0392-100X-N2336","RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"OTORHINOLARYNGOLOGY","Score":null,"Total":0}

引用次数: 0

Abstract

Objective: To achieve instance segmentation of upper aerodigestive tract (UADT) neoplasms using a deep learning (DL) algorithm, and to identify differences in its diagnostic performance in three different sites: larynx/hypopharynx, oral cavity and oropharynx.

Methods: A total of 1034 endoscopic images from 323 patients were examined under narrow band imaging (NBI). The Mask R-CNN algorithm was used for the analysis. The dataset split was: 935 training, 48 validation and 51 testing images. Dice Similarity Coefficient (Dsc) was the main outcome measure.

Results: Instance segmentation was effective in 76.5% of images. The mean Dsc was 0.90 ± 0.05. The algorithm correctly predicted 77.8%, 86.7% and 55.5% of lesions in the larynx/hypopharynx, oral cavity, and oropharynx, respectively. The mean Dsc was 0.90 ± 0.05 for the larynx/hypopharynx, 0.60 ± 0.26 for the oral cavity, and 0.81 ± 0.30 for the oropharynx. The analysis showed inferior diagnostic results in the oral cavity compared with the larynx/hypopharynx (p < 0.001).

Conclusions: The study confirms the feasibility of instance segmentation of UADT using DL algorithms and shows inferior diagnostic results in the oral cavity compared with other anatomic areas.

Abstract Image

查看原文本刊更多论文

上消化道癌症的实例分割:部位特异性结果。

目的:利用深度学习(DL)算法实现上气消化道(UADT)肿瘤的实例分割，并识别其在喉/下咽、口腔和口咽三个不同部位的诊断性能差异。方法:对323例患者的1034张内镜图像进行窄带成像(NBI)检查。使用Mask R-CNN算法进行分析。数据集分割为:935个训练图像，48个验证图像和51个测试图像。骰子相似系数(Dsc)是主要的结局指标。结果:实例分割的准确率为76.5%。平均Dsc为0.90±0.05。该算法对喉/下咽、口腔和口咽部病变的正确率分别为77.8%、86.7%和55.5%。喉/下咽的平均Dsc为0.90±0.05，口腔为0.60±0.26，口咽部为0.81±0.30。分析显示口腔的诊断结果较喉/下咽差(p < 0.001)。结论:本研究证实了使用DL算法对UADT进行实例分割的可行性，但在口腔的诊断效果较其他解剖区域差。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Acta Otorhinolaryngologica Italica OTORHINOLARYNGOLOGY-

CiteScore

3.40

自引率

10.00%

发文量

审稿时长

6-12 weeks

期刊介绍： Acta Otorhinolaryngologica Italica first appeared as “Annali di Laringologia Otologia e Faringologia” and was founded in 1901 by Giulio Masini. It is the official publication of the Italian Hospital Otology Association (A.O.O.I.) and, since 1976, also of the Società Italiana di Otorinolaringoiatria e Chirurgia Cervico-Facciale (S.I.O.Ch.C.-F.). The journal publishes original articles (clinical trials, cohort studies, case-control studies, cross-sectional surveys, and diagnostic test assessments) of interest in the field of otorhinolaryngology as well as clinical techniques and technology (a short report of unique or original methods for surgical techniques, medical management or new devices or technology), editorials (including editorial guests – special contribution) and letters to the Editor-in-Chief. Articles concerning science investigations and well prepared systematic reviews (including meta-analyses) on themes related to basic science, clinical otorhinolaryngology and head and neck surgery have high priority.