In the 2023 World Report on Vision, the World Health Organization reports that at least 2.2 billion people worldwide currently suffer from serious eye disorders. Of these, over one billion have a preventable blindness condition, a term used to refer to problems such as cataracts, retinopathy, diabetic macular edema or glaucoma, which could be managed or prevented through early diagnosis and intervention, Thus reducing visual impairment. In Italy, a report by the Ministry of Health shows that 19% of the population has severe or moderate eye problems, with 34% over 65 and 42% over 75. The onset of certain eye diseases can make it impossible to perform even mundane daily activities such as reading, walking or shopping. In less developed countries, visual disabilities not only compromise autonomy and mobility but also increase the risk of social and work isolation as well as mortality. The consequences are particularly serious for children, as vision problems during the developmental age may limit neuro-psychomotor learning and development. Despite this, according to the Ministry of Health, most children undergo their first screening at around seven years old, while screening in the three- and four-year age group, which is crucial for early diagnosis, is still not widespread, although they are essential to prevent the onset of any disease. In addition, neonatal centres are increasingly required to perform retinal screening of infants, a task for which there is no specific tool to assist this important preventive action effectively. In the field of ophthalmology and vision prevention, eyes.guru srl is active, a Padova start-up born in November 2023, committed to developing intelligent solutions for the diagnosis of eye diseases. Here I did a six-month internship, helping to design an innovative device for detecting defects in the human eye through imaging techniques. Cameras with the same lens currently on the market use cameras with light sources used to capture a static image on various dials. Thanks to a particular combination of LEDs that will be described in the thesis, the device eyes.guru obtains for each patient a database of more qualified images, which carry a huge amount of information dependent on the sequences of lighting lights, their positions and intensity. This innovative system allows simplified automatic detection (already patented) and provides an objective response without requiring collaboration from the patient: it does not matter if he is a newborn, If there is a language barrier between the elderly person and the operator who is taking the images, it will be sufficient for the patient to look at the device at one point or another in order to obtain all the information needed to diagnose any visual defects. My tasks during the period spent at eyes.guru were three: 1) Analytically analyze a significant series of images collected by a test eye, extracting the most relevant parameters to determine uniquely the refractive defect associated with the lens in front of the camera. 2) Train a machine learning algorithm by using the gray-level light intensity profiles of thousands of images from the test eye. 3) Compare the results obtained from machine learning with the results obtained from analytical analysis. The findings are illustrated in the following thesis.
Nel World Report On Vision del 2023, l’Organizzazione Mondiale della Sanità segnala che attualmente nel mondo almeno 2,2 miliardi di persone soffrono di gravi disturbi oculari. Di queste, oltre un miliardo ha una condizione di cecità evitabile, termine utilizzato per indicare problemi come cataratta, retinopatia, edema maculare diabetico o glaucoma, che potrebbero essere gestiti o prevenuti grazie a diagnosi precoci e interventi tempestivi, riducendo così la disabilità visiva. In Italia, una relazione del ministero della Salute evidenzia che il 19% della popolazione presenta problemi oculari gravi o moderati, percentuale che sale al 34% tra gli over 65 e al 42% tra gli over 75. L’insorgenza di alcune patologie oculari può rendere impossibile svolgere anche banali attività quotidiane come leggere, camminare o fare la spesa. Nei paesi meno sviluppati, le disabilità visive compromettono non solo l’autonomia e la mobilità, ma aumentano anche il rischio di isolamento sociale e lavorativo, oltre che di mortalità. Le conseguenze sono particolarmente gravi anche per i bambini, poiché i problemi visivi durante l’età evolutiva possono limitare l’apprendimento e lo sviluppo neuro-psicomotorio. Nonostante ciò, secondo il ministero della Salute, la maggior parte dei bambini effettua il primo controllo intorno ai sette anni, mentre gli screening nella fascia d’età tra i tre e i quattro anni, cruciali per la diagnosi precoce, sono ancora poco diffusi, sebbene siano essenziali per prevenire l'insorgenza di eventuali patologie. Inoltre, ai centri neonatali viene sempre più richiesto di eseguire uno screening retinico ai neonati, incarico per il quale manca uno strumento specifico che aiuti in questa importante azione preventiva in modo efficace. Proprio nel campo dell’oftalmologia e della prevenzione visiva è attiva eyes.guru srl, start-up padovana nata nel novembre 2023, impegnata nello sviluppo di soluzioni intelligenti per la diagnosi di patologie oculari. Qui ho svolto un tirocinio di sei mesi, contribuendo alla progettazione di un dispositivo innovativo per la rilevazione dei difetti dell’occhio umano attraverso tecniche di imaging. Gli apparecchi con lo stesso obiettivo attualmente sul mercato utilizzano fotocamere con delle sorgenti di luce utilizzate per acquisire un’immagine statica su vari quadranti. Grazie a una particolare combinazione di LED che verrà descritta nella tesi, il dispositivo di eyes.guru ricava per ogni paziente un database di immagini più qualificate, che portano una grandissima quantità di informazioni dipendenti dalle sequenze di accensione delle luci, dalle loro posizioni e intensità. Questo sistema innovativo consente una rilevazione automatica semplificata (già brevettata) e fornisce una risposta oggettiva senza richiedere collaborazione al paziente: non importa che egli sia un neonato, un anziano o che vi sia una barriera linguistica con l’operatore che sta eseguendo la raccolta delle immagini, basterà che il paziente, in un momento o in un altro, guardi verso il dispositivo per poter ricavare tutte le informazioni utili alla diagnosi di eventuali difetti visivi. I miei compiti nel periodo trascorso ad eyes.guru sono stati tre: 1) Analizzare analiticamente una significativa serie di immagini raccolte tramite un occhio di prova, estraendo i parametri più rilevanti per determinare in maniera univoca il difetto refrattivo associato alla lente posta davanti alla telecamera. 2) Eseguire il training di un algoritmo di machine learning tramite i profili di intensità di luce a livelli di grigio di migliaia di immagini ottenute dall’occhio di prova. 3) Confrontare i risultati ottenuti dal machine learning con i risultati ottenuti dall’analisi analitica. Quanto è emerso viene illustrato nel seguente lavoro di tesi.
Nuove Tecnologie per la Diagnosi Visiva: Sviluppo di un Dispositivo Basato su illuminazione a LED e Machine Learning
ROSSI, LORENZO
2023/2024
Abstract
In the 2023 World Report on Vision, the World Health Organization reports that at least 2.2 billion people worldwide currently suffer from serious eye disorders. Of these, over one billion have a preventable blindness condition, a term used to refer to problems such as cataracts, retinopathy, diabetic macular edema or glaucoma, which could be managed or prevented through early diagnosis and intervention, Thus reducing visual impairment. In Italy, a report by the Ministry of Health shows that 19% of the population has severe or moderate eye problems, with 34% over 65 and 42% over 75. The onset of certain eye diseases can make it impossible to perform even mundane daily activities such as reading, walking or shopping. In less developed countries, visual disabilities not only compromise autonomy and mobility but also increase the risk of social and work isolation as well as mortality. The consequences are particularly serious for children, as vision problems during the developmental age may limit neuro-psychomotor learning and development. Despite this, according to the Ministry of Health, most children undergo their first screening at around seven years old, while screening in the three- and four-year age group, which is crucial for early diagnosis, is still not widespread, although they are essential to prevent the onset of any disease. In addition, neonatal centres are increasingly required to perform retinal screening of infants, a task for which there is no specific tool to assist this important preventive action effectively. In the field of ophthalmology and vision prevention, eyes.guru srl is active, a Padova start-up born in November 2023, committed to developing intelligent solutions for the diagnosis of eye diseases. Here I did a six-month internship, helping to design an innovative device for detecting defects in the human eye through imaging techniques. Cameras with the same lens currently on the market use cameras with light sources used to capture a static image on various dials. Thanks to a particular combination of LEDs that will be described in the thesis, the device eyes.guru obtains for each patient a database of more qualified images, which carry a huge amount of information dependent on the sequences of lighting lights, their positions and intensity. This innovative system allows simplified automatic detection (already patented) and provides an objective response without requiring collaboration from the patient: it does not matter if he is a newborn, If there is a language barrier between the elderly person and the operator who is taking the images, it will be sufficient for the patient to look at the device at one point or another in order to obtain all the information needed to diagnose any visual defects. My tasks during the period spent at eyes.guru were three: 1) Analytically analyze a significant series of images collected by a test eye, extracting the most relevant parameters to determine uniquely the refractive defect associated with the lens in front of the camera. 2) Train a machine learning algorithm by using the gray-level light intensity profiles of thousands of images from the test eye. 3) Compare the results obtained from machine learning with the results obtained from analytical analysis. The findings are illustrated in the following thesis.File | Dimensione | Formato | |
---|---|---|---|
ROSSI_LORENZO.pdf
embargo fino al 12/12/2027
Dimensione
5.46 MB
Formato
Adobe PDF
|
5.46 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/79759