Histopathology, the microscopic analysis of biological tissues, is one of the main tools for disease diagnosis. Traditionally, this analysis has relied only on manual image evaluation, an extensive and subjective process, affected by observer variability and possible human errors. In recent years, artificial intelligence has brought significant innovations to the field of digital pathology. Deep learning algorithms have been developed for the processing and automatic classification of digitized histological images through Whole Slide Imaging (WSI) systems. These automated decision-making systems can support physicians in achieving faster, more accurate, and more objective diagnoses. This thesis introduces some key concepts, such as histopathology and its digitization, machine learning, deep learning, and neural networks, as well as the central role of WSI images in predictive models. It then explores convolutional neural networks (CNNs) in the context of automated histopathological diagnosis, from the training phase, to image processing, and classification. Main public datasets and the interpretability of algorithms are also examined. The thesis ends by presenting the advantages, limitations and future prospects of artificial intelligence in the field of histopathological diagnostics.
L’istopatologia, ovvero l’analisi microscopica di tessuti biologici, è uno degli strumenti principali per la diagnosi di patologie. Questa analisi si è basata in passato sulla sola valutazione manuale delle immagini, processo lungo e soggettivo, influenzato da variabilità dell’osservatore e soggetto a possibili errori umani. Negli ultimi anni, l’intelligenza artificiale ha portato a diverse innovazioni nell’ambito della patologia digitale. Sono stati sviluppati algoritmi di deep learning per l’elaborazione e classificazione automatica di immagini istologiche digitalizzate tramite sistemi di Whole Slide Imaging (WSI). Tali sistemi decisionali automatici possono affiancare il medico, per una diagnosi più rapida, accurata e imparziale. Questa tesi introduce alcuni concetti fondamentali quali l’istopatologia e la sua digitalizzazione, machine learning, deep learning e reti neurali, oltre al ruolo centrale delle immagini WSI nei modelli predittivi. Vengono poi approfondite le reti neurali convoluzionali (CNN) nell’ambito della diagnostica istopatologica automatizzata, dalla fase di training, all’elaborazione e classificazione delle immagini. Sono inoltre valutati i principali dataset disponibili e l’interpretabilità degli algoritmi. La tesi si conclude esponendo vantaggi, limiti e prospettive future dell'intelligenza artificiale nell'ambito della diagnostica istopatologica.
Deep learning e immagini istopatologiche: verso la diagnosi automatizzata
MACALUSO, GIULIA
2024/2025
Abstract
Histopathology, the microscopic analysis of biological tissues, is one of the main tools for disease diagnosis. Traditionally, this analysis has relied only on manual image evaluation, an extensive and subjective process, affected by observer variability and possible human errors. In recent years, artificial intelligence has brought significant innovations to the field of digital pathology. Deep learning algorithms have been developed for the processing and automatic classification of digitized histological images through Whole Slide Imaging (WSI) systems. These automated decision-making systems can support physicians in achieving faster, more accurate, and more objective diagnoses. This thesis introduces some key concepts, such as histopathology and its digitization, machine learning, deep learning, and neural networks, as well as the central role of WSI images in predictive models. It then explores convolutional neural networks (CNNs) in the context of automated histopathological diagnosis, from the training phase, to image processing, and classification. Main public datasets and the interpretability of algorithms are also examined. The thesis ends by presenting the advantages, limitations and future prospects of artificial intelligence in the field of histopathological diagnostics.| File | Dimensione | Formato | |
|---|---|---|---|
|
Macaluso_Giulia.pdf
accesso aperto
Dimensione
951.92 kB
Formato
Adobe PDF
|
951.92 kB | Adobe PDF | Visualizza/Apri |
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/20.500.12608/92562