Integrative Digital Pathology for Personalized Medicine: Population stratification and early biomarkers findings, consolidating and completing the use of Prostate-Specific Antigen (PSA) and Gleason Score in Prostate Cancer
No Thumbnail Available
Date
2025-04-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Pontificia Universidad Católica del Perú
Acceso al texto completo solo para la Comunidad PUCP
Abstract
biomarcadores establecidos para predecir el riesgo de recaída y el riesgo de presentar
recurrencia bioquímica. Una comprensión más profunda del comportamiento de los
tejidos proporcionada por técnicas moleculares puede mejorar la capacidad de
pronosticar la probabilidad de recurrencia. Basándose en datos sólidos y correctamente
anotados disponibles de grandes cohortes internacionales de pacientes, y en el
procesamiento exhaustivo de datos de información fenotípica y genómica, este trabajo
propuso y evaluó el papel de los biomarcadores tempranos de recurrencia. Además, se
incluyó en el análisis información clínica asociada a datos estructurales y moleculares
del tejido para proporcionar una comprensión más profunda del microentorno del cáncer
de próstata. Por lo tanto, se entrenaron modelos de aprendizaje profundo para segmentar
características morfológicas de imágenes de diapositivas completas, descargadas de
repositorios disponibles públicamente. Las características segmentadas estaban
asociadas a la proliferación celular, la estructura de la luz y la arquitectura de la región
tumoral. A continuación, se predijo el riesgo de presentar recurrencia se predijo
entonces mediante algoritmos de aprendizaje automático a partir de las características
tisulares mencionadas, y se analizó el papel de la puntuación de Gleason. Al mismo
tiempo, se introdujeron en los modelos niveles de expresión genómica pre-procesados
para recuperar un subconjunto de genes responsables de la recurrencia. Los resultados
indican que, tras la inspección de los biomarcadores, la organización de la matriz
extracelular se ha asociado con el riesgo de presentar recurrencia. Además, se
establecieron los niveles de PSA como información crítica a la hora de detectar la
recurrencia. Los algoritmos de aprendizaje automático entrenados en el genoma y el
fenotipo clasificaron a los pacientes con una precisión media del 79 % y el 69,7 %,
respectivamente, cuando la recurrencia bioquímica se produjo hasta 22 meses después
de su tratamiento final, lo que demuestra que el riesgo de presentar recurrencia
bioquímica puede predecirse con éxito cuando se integra la información clínica,
fenotípica y genómica.
Prostate cancer, although the second most common cancer in men, does not have established biomarkers to predict the risk of relapse and risk of presenting biochemical recurrence. A deeper comprehension of tissue behavior provided by molecular techniques may enhance the ability to forecast the likelihood of recurrence. Based on robust and highly annotated data available from large international cohorts of patients, and extensive data processing of phenotypic and genomic information, this work proposed and evaluated the role of early biomarkers of recurrence. Furthermore, clinical information associated with structural and molecular data from the tissue was included in the analysis to provide a deeper understanding of the prostate cancer microenvironment. Deep learning models were hence trained to segment morphological features from Whole Slide Images, downloaded from publicly available repositories. Features segmented were associated with cell proliferation, lumen structure, and tumorous region architecture. The risk of presenting recurrence was then predicted via machine learning algorithms from the aforementioned tissue features, and the role of the Gleason score was analyzed. Concurrently, pre-processed genomic levels of expression were fed to models to retrieve a subset of genes responsible for the recurrence. The results indicate that after inspection of biomarkers, extracellular matrix organization has been associated with the risk of presenting recurrence. In addition, PSA levels were established as critical information when detecting recurrence. Machine learning algorithms trained on the genome and phenotype classified patients with an average precision of 79% and 69.7%, respectively, when the biochemical recurrence occurred up to 22 months after its final treatment, providing evidence that the risk of presenting biochemical recurrence can be successfully predicted when clinical, phenotypic and genomic information are integrated.
Prostate cancer, although the second most common cancer in men, does not have established biomarkers to predict the risk of relapse and risk of presenting biochemical recurrence. A deeper comprehension of tissue behavior provided by molecular techniques may enhance the ability to forecast the likelihood of recurrence. Based on robust and highly annotated data available from large international cohorts of patients, and extensive data processing of phenotypic and genomic information, this work proposed and evaluated the role of early biomarkers of recurrence. Furthermore, clinical information associated with structural and molecular data from the tissue was included in the analysis to provide a deeper understanding of the prostate cancer microenvironment. Deep learning models were hence trained to segment morphological features from Whole Slide Images, downloaded from publicly available repositories. Features segmented were associated with cell proliferation, lumen structure, and tumorous region architecture. The risk of presenting recurrence was then predicted via machine learning algorithms from the aforementioned tissue features, and the role of the Gleason score was analyzed. Concurrently, pre-processed genomic levels of expression were fed to models to retrieve a subset of genes responsible for the recurrence. The results indicate that after inspection of biomarkers, extracellular matrix organization has been associated with the risk of presenting recurrence. In addition, PSA levels were established as critical information when detecting recurrence. Machine learning algorithms trained on the genome and phenotype classified patients with an average precision of 79% and 69.7%, respectively, when the biochemical recurrence occurred up to 22 months after its final treatment, providing evidence that the risk of presenting biochemical recurrence can be successfully predicted when clinical, phenotypic and genomic information are integrated.
Description
Keywords
Próstata--Cáncer, Biomarcadores, Aprendizaje automático (Inteligencia artificial), Fenotipo, Genoma humano
Citation
Collections
Endorsement
Review
Supplemented By
Referenced By
Creative Commons license
Except where otherwised noted, this item's license is described as info:eu-repo/semantics/openAccess