Medical image segmentation is a valuable procedure in many healthcare applications, requiring reliable methods to evaluate algorithmic performance. This thesis presents a customizable web application that enables users to interactively explore segmentation performance metrics – such as Dice similarity coefficient (DSC) and Hausdorff distance (HD) – in a flexible and visual environment. The application is composed of three layers: (1) a RESTful API built in Node.js and TypeScript, employing a functional approach that adheres to the single responsibility principle while emphasizing modularity, maintainability, and security; (2) a NoSQL database implemented with MongoDB; and (3) a fully data-driven interface developed in React and TypeScript. Together, these layers integrate user-defined Scalable Vector Graphics (SVG) with user-supplied data to deliver a customized analytics experience, enabling dynamic data-driven visualizations, custom filters, and statistics. The open-source repository includes a continuous integration and continuous deployment (CI/CD) pipeline that automates deployment to both a virtual private server and a Docker hub repository, allowing robust version control and reproducibility. By adopting a user-driven approach to segmentation analytics, this system empowers researchers and clinicians to gain deeper, actionable insights into medical image segmentation performance. The web application is publicly accessible at https://www.mis-analytics.org along with the public Docker images available at https://hub.docker.com/r/nicolasleivab/mis-analytics and https://hub.docker.com/r/nicolasleivab/mis-analytics-service.
La segmentazione delle immagini mediche è una procedura preziosa in molte applicazioni sanitarie, che richiede metodi affidabili per valutare le prestazioni degli algoritmi. Questa tesi presenta un'applicazione web personalizzabile che consente agli utenti di esplorare interattivamente le metriche di segmentazione - come il coefficiente Dice (DSC) e la distanza di Hausdorff (HD) - in un ambiente flessibile e visivo. L'applicazione è composta da tre livelli: (1) un'API RESTful costruita in Node.js e TypeScript, utilizzando un approccio funzionale che aderisce al principio della singola responsabilità, enfatizzando al contempo modularità, manutenibilità e sicurezza; (2) un database NoSQL implementato con MongoDB; e (3) un'interfaccia completamente data-driven sviluppata in React e TypeScript. Insieme, questi livelli integrano gli Scalable Vector Graphics (SVG) definiti dall'utente con i dati forniti dall'utente per offrire un'esperienza analitica personalizzata, consentendo visualizzazioni dinamiche basate sui dati, filtri personalizzati e statistiche. Il repository open-source include una pipeline di integrazione e distribuzione continua (CI/CD) che automatizza la distribuzione sia su un server privato virtuale che su un repository hub Docker, consentendo un solido controllo delle versioni e la riproducibilità. Adottando un approccio all'analisi della segmentazione guidato dall'utente, questo sistema consente ai ricercatori e ai medici di ottenere informazioni più profonde e fruibili sulle prestazioni della segmentazione delle immagini mediche. L'applicazione web è accessibile pubblicamente all'indirizzo https://www.mis-analytics.org, così come le immagini Docker pubbliche disponibili su https://hub.docker.com/r/nicolasleivab/mis-analytics e https://hub.docker.com/r/nicolasleivab/mis-analytics-service.
A customizable web application for enhancing medical image segmentation analysis
Leiva Büchi, Nicolás Andrés
2023/2024
Abstract
Medical image segmentation is a valuable procedure in many healthcare applications, requiring reliable methods to evaluate algorithmic performance. This thesis presents a customizable web application that enables users to interactively explore segmentation performance metrics – such as Dice similarity coefficient (DSC) and Hausdorff distance (HD) – in a flexible and visual environment. The application is composed of three layers: (1) a RESTful API built in Node.js and TypeScript, employing a functional approach that adheres to the single responsibility principle while emphasizing modularity, maintainability, and security; (2) a NoSQL database implemented with MongoDB; and (3) a fully data-driven interface developed in React and TypeScript. Together, these layers integrate user-defined Scalable Vector Graphics (SVG) with user-supplied data to deliver a customized analytics experience, enabling dynamic data-driven visualizations, custom filters, and statistics. The open-source repository includes a continuous integration and continuous deployment (CI/CD) pipeline that automates deployment to both a virtual private server and a Docker hub repository, allowing robust version control and reproducibility. By adopting a user-driven approach to segmentation analytics, this system empowers researchers and clinicians to gain deeper, actionable insights into medical image segmentation performance. The web application is publicly accessible at https://www.mis-analytics.org along with the public Docker images available at https://hub.docker.com/r/nicolasleivab/mis-analytics and https://hub.docker.com/r/nicolasleivab/mis-analytics-service.| File | Dimensione | Formato | |
|---|---|---|---|
|
Leiva_Büchi_Thesis_Corrections.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Thesis
Dimensione
5.13 MB
Formato
Adobe PDF
|
5.13 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/234901