Quantification and evaluation of the diagnostic significance of adenocarcinoma-associated microenvironmental changes in the prostate using modern digital pathology solutions
Date
2021-10-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Eesnäärme adenokartsinoom on meestel diagnoositud pahaloomulistest kasvajatest maailmas sageduselt teisel kohal ning viiendal kohal vähkidest põhjustatud meeste surmadest. Tervise Arengu Instituudi andmetel diagnoositi 2017. aastal Eestis prostata adenokartsinoomi 1113 juhul, mis moodustas 25,2% kõikidest pahaloomulistest kasvajatest. Haiguse diagnoos püstitatakse enne radikaalset prostatektoomiat kasutades biopsiate uuringut. Kuigi eesnäärme bioptaatide käsitluses toimub pidev areng, on jätkuvalt suur tõenäosus, et radikaalse prostatektoomia järgselt muutub kasvaja histoloogiline aste – Gleason’i skoor hinnatakse prostatektioomia materjali alusel raskemaks 23,3% kuni 42,7% juhtudest.
Kõikidel bioptaatidel on epiteliaalset komponenti ümbritsev mikrokeskkond, mis potentsiaalselt võib anda täiendavat diagnostilist ja prognostilist informatsiooni. Selle hüpoteesi testimiseks kasutati kahte stromaalset markerit: Masson’i trikroomi ja antud paikme vaates uut markerit anti-FANCM antikeha.
Kvantitatiivne stromaalsete muutuste hindamine mikroskoobis on sageli aeganõudev, väljakutseid pakkuv ning hindajast lähtuvalt subjektiivne. Tänapäevaste digitaalse patoloogia lahendustega saaks hinnangu anda kiirelt ja usaldusväärselt. Uurimistöö käigus töötati välja avatud lähtekoodiga programm Pathadin, mis imiteerib patoloogi üldiseid töövõtteid. Arsti poolt treenitud mudel õppis eristama eesnäärme erinevaid struktuure – näärmeid, närve, stroomat, rasva ja optilisi tühimikke, analüüsima neid, kasutama erinevaid filtreid, näiteks värvifiltrit strooma analüüsiks (FANCM, Masson’i trikroom) ning loendama näärmelises komponendis DAB positiivsete rakkude hulka. Mudel sai hakkama oluliste diagnostiliste parameetrite skriinimisega, näiteks tuvastas eesnäärmes edukalt perineuraalse invasiooni ja ekstraprostaatilise leviku.
Uurimistöö tulemuste alusel saab välja tuua mitmed olulised järeldused. Esiteks kirjeldab töö süsteemselt ära uue immunohistokeemilise markeri FACNM kasutamise eesnäärmes. Teiseks tõestab, et stromaalset päritolu muutused on Gleason’i skoorist sõltuvad, kuid on piiratud väärtusega madala tundlikkuse tõttu. Kolmandaks näitab, et masinõpe kui mitte veel täielikult usaldusväärne inimesest sõltumatute lõplike diagnooside püstitamisel, võib juba praegu olla abiks histoloogilistes ja teaduslikes uuringutes.
Töö kirjelduses esitatud juhendid on universaalsed: neid saab kasutada erinevatel kudedel, nad võiksid julgustada patolooge kasutama arvuti poolt abistatavat diagnostikat ning looma ja arendama oma mudeleid.
Prostate adenocarcinoma is the second most frequently diagnosed cancer in males worldwide. According to 2017 data of the Estonian National Institute for Health Development, it was diagnosed in 1113 cases, making 25.2% of all diagnosed malignancies. The standard for the definitive diagnosis prior to the radical prostatectomy is a systematic biopsy sampling. Nevertheless, despite the progress in biopsy sampling, the final Gleason score is upgraded in 23.3% to 42.7% of all radical prostatectomy samples. To increase the concordance between biopsy and prostatectomy, an idea of evaluating the diagnostic significance of microenvironmental changes surrounding the epithelial component has emerged. Two markers were tested to study stromogenic changes — Masson’s trichrome and, a novel in the prostate, anti-FANCM antibody. The precise quantification of histological stains using a microscope is frequently time-consuming, challenging, and depends on human reliability. However, modern digital pathology solutions could perform the analysis quickly and accurately. During the studies, an open-source set of tools under the name of Pathadin and a model for prostate segmentation were developed. The program imitates pathologists' work in its basics: trained by a doctor, it learned to distinguish glands, nerves, stroma, fat, and empty compartments in the prostate to analyze these independently and apply specific filters such as color analysis for FANCM and Masson’s trichrome in the stroma, or DAB positive cell counting in the glandular component. The model was also able to assist in the screening of significant diagnostic features, as the perineural invasion or extraprostatic extension. The work had several essential outcomes. Firstly, a novel in the prostate immunohistochemical marker, FACNM, was systematically described. Secondly, it was shown that stromogenic changes are indeed Gleason dependant yet are of a limited clinical value due to low sensitivity. Thirdly, if not yet fully reliable for independent definitive diagnoses, machine learning can already be beneficial in histological routines and scientific research. The workflow and manuals provided in the manuscript are somewhat universal and can be used for different tissues, encouraging pathologists to test computer-assisted diagnostics and train their own, more advanced, models.
Prostate adenocarcinoma is the second most frequently diagnosed cancer in males worldwide. According to 2017 data of the Estonian National Institute for Health Development, it was diagnosed in 1113 cases, making 25.2% of all diagnosed malignancies. The standard for the definitive diagnosis prior to the radical prostatectomy is a systematic biopsy sampling. Nevertheless, despite the progress in biopsy sampling, the final Gleason score is upgraded in 23.3% to 42.7% of all radical prostatectomy samples. To increase the concordance between biopsy and prostatectomy, an idea of evaluating the diagnostic significance of microenvironmental changes surrounding the epithelial component has emerged. Two markers were tested to study stromogenic changes — Masson’s trichrome and, a novel in the prostate, anti-FANCM antibody. The precise quantification of histological stains using a microscope is frequently time-consuming, challenging, and depends on human reliability. However, modern digital pathology solutions could perform the analysis quickly and accurately. During the studies, an open-source set of tools under the name of Pathadin and a model for prostate segmentation were developed. The program imitates pathologists' work in its basics: trained by a doctor, it learned to distinguish glands, nerves, stroma, fat, and empty compartments in the prostate to analyze these independently and apply specific filters such as color analysis for FANCM and Masson’s trichrome in the stroma, or DAB positive cell counting in the glandular component. The model was also able to assist in the screening of significant diagnostic features, as the perineural invasion or extraprostatic extension. The work had several essential outcomes. Firstly, a novel in the prostate immunohistochemical marker, FACNM, was systematically described. Secondly, it was shown that stromogenic changes are indeed Gleason dependant yet are of a limited clinical value due to low sensitivity. Thirdly, if not yet fully reliable for independent definitive diagnoses, machine learning can already be beneficial in histological routines and scientific research. The workflow and manuals provided in the manuscript are somewhat universal and can be used for different tissues, encouraging pathologists to test computer-assisted diagnostics and train their own, more advanced, models.
Description
Väitekirja elektrooniline versioon ei sisalda publikatsioone
Keywords
diseases of prostate, prostatic adenocarcinoma, biopsy, histopathological diagnostics, computer aided diagnosis