Choroby siatkówki
Rokowania, prognozy i postęp choroby

Choroby siatkówki, takie jak dziedziczne choroby siatkówki (IRD) oraz zwyrodnienie plamki żółtej związane z wiekiem (AMD), stanowią główną przyczynę utraty wzroku w krajach rozwiniętych. Wczesne rozpoznanie i właściwa diagnoza są kluczowe dla zapobiegania trwałej utracie widzenia, a nowoczesne metody leczenia i monitorowania pozwalają na spowolnienie progresji choroby. W szczególności, modele oparte na uczeniu głębokim (deep learning, DL) i sztucznej inteligencji (AI) wykazały wysoką skuteczność prognostyczną, np. w przewidywaniu ryzyka rozwoju późnego AMD z 5-letnią C-statystyką na poziomie 86,4%, czy w klasyfikacji genów przyczynowych IRD z dokładnością powyżej 80%. W retinitis pigmentosa (RP) algorytmy DL umożliwiają ocenę ostrości wzroku z AUC 0,85, co pozwala na precyzyjne prognozowanie funkcji wzrokowej na podstawie obrazów OCT. Ponadto, zaawansowane modele, takie jak Longitudinalny Transformer do Analizy Przeżycia (LTSA), umożliwiają dynamiczne prognozowanie chorób na podstawie sekwencji obrazów dna oka, przewyższając modele oparte na pojedynczych obrazach.

  1. Prognoza chorób siatkówki (Choroby siatkówki – rokowanie)
    1. Znaczenie prognozowania chorób siatkówki
    2. Wykorzystanie uczenia głębokiego w prognozowaniu chorób siatkówki
    3. Innowacyjne modele prognozowania
    4. Biomarkery prognostyczne w obrazowaniu siatkówki
    5. Skuteczność prognozowania chorób siatkówki
    6. Ograniczenia obecnych metod prognozowania
    7. Przyszłość prognozowania chorób siatkówki
  2. Znaczenie kliniczne prognozowania chorób siatkówki
    1. Kolejne rozdziały

Prognoza chorób siatkówki (Choroby siatkówki – rokowanie)

Choroby siatkówki stanowią główną przyczynę utraty wzroku w krajach rozwiniętych, mając największy udział w upośledzeniu widzenia u dzieci, osób w wieku produkcyjnym (dziedziczne choroby siatkówki) oraz osób starszych (zwyrodnienie plamki żółtej związane z wiekiem – AMD).1 Wczesne rozpoznanie tych chorób i właściwa diagnoza mogą zapobiec trwałej utracie wzroku. Dzięki odpowiedniemu leczeniu i systematycznemu monitorowaniu możliwe jest spowolnienie lub zahamowanie dalszego pogorszenia widzenia, szczególnie gdy schorzenie zostanie wykryte w początkowej fazie.2 Obecnie istnieje wiele innowacyjnych metod pozwalających na przewidywanie przebiegu chorób siatkówki, co ma kluczowe znaczenie dla podejmowania decyzji klinicznych, efektywnego planowania leczenia i monitorowania pacjentów.

Znaczenie prognozowania chorób siatkówki

Dokładne przewidywanie progresji chorób siatkówki ma kluczowe znaczenie kliniczne. Umożliwia to podejmowanie lepszych decyzji dotyczących: (1) leczenia farmakologicznego, szczególnie suplementów doustnych znanych z ograniczania ryzyka progresji, (2) interwencji dotyczących stylu życia, zwłaszcza zaprzestania palenia tytoniu i zmian dietetycznych, oraz (3) intensywności monitorowania pacjenta, np. częstych badań kontrolnych w klinice i/lub dostosowanych programów monitorowania domowego.3 W przypadku rzadkich chorób postawienie ostatecznej diagnozy może zająć kilka lat (tzw. odyseja diagnostyczna), co prowadzi do niepewności co do rokowania i opóźnienia w zapewnieniu odpowiedniej opieki.4

Dla wolno postępujących chorób oczu, takich jak zwyrodnienie plamki żółtej związane z wiekiem (AMD) i jaskra pierwotna otwartego kąta (POAG), pacjenci przechodzą powtarzające się badania obrazowe w czasie, aby śledzić postęp choroby. Prognozowanie przyszłego ryzyka rozwoju choroby jest kluczowe dla właściwego planowania leczenia.5 W przypadku obu tych chorób, dokładne zidentyfikowanie pacjentów wysokiego ryzyka tak wcześnie, jak to możliwe, ma decydujące znaczenie dla podejmowania decyzji klinicznych, pomagając w planowaniu postępowania, leczenia lub monitorowania pacjenta.6

Wykorzystanie uczenia głębokiego w prognozowaniu chorób siatkówki

Uczenie głębokie (deep learning, DL) stało się przełomem w zautomatyzowanej diagnostyce na podstawie obrazowania medycznego, z wieloma udanymi zastosowaniami w okulistyce.7 W ostatnich latach algorytmy wykorzystujące sztuczną inteligencję (AI) znacząco poprawiły możliwości przewidywania przebiegu chorób siatkówki:

  • Zwyrodnienie plamki żółtej związane z wiekiem (AMD) – badania AREDS i AREDS2 wykorzystały algorytmy DL i analizę przeżycia do przewidywania ryzyka rozwoju późnego AMD, osiągając wysoką dokładność prognostyczną.8 Modele DL połączone z technikami analizy przeżycia osiągnęły wysoką dokładność prognostyczną (5-letni wskaźnik C-statystyki 86,4), która znacznie przewyższała oceny specjalistów siatkówki wykorzystujących dwa istniejące standardy kliniczne.910
  • Dziedziczne choroby siatkówki (IRD) – opracowano algorytmy AI wykorzystujące multimodalne techniki obrazowania do ułatwienia diagnozy, klasyfikacji, odkrywania etiologii genetycznej i pomiaru tempa progresji IRD.11 Badania wykazały możliwość przewidywania genów przyczynowych w IRD na podstawie fotografii dna oka i obrazowania autofluorescencji dna oka (FAF) z wysoką dokładnością przekraczającą 80%.12
  • Retinitis pigmentosa (RP) – ocena przewidywania ostrości wzroku na podstawie obrazów OCT i podczerwieni u pacjentów z RP wykazała możliwość określenia, czy pacjent ma ostrość wzroku poniżej lub powyżej 20/40, z AUC wynoszącym 0,85.13 Algorytm DL może rozróżniać między dwoma poziomami ostrości wzroku z relatywnie wysoką czułością i swoistością, wykorzystując tylko pojedynczy przekrój transfowealny OCT jako dane wejściowe.14

Innowacyjne modele prognozowania

Najnowsze badania w dziedzinie prognozowania chorób siatkówki skupiają się na rozwoju zaawansowanych modeli prognostycznych:

  • Model Longitudinalnego Transformera do Analizy Przeżycia (LTSA) – umożliwia dynamiczne prognozowanie chorób na podstawie wzdłużnego obrazowania medycznego, modelując czas do wystąpienia choroby na podstawie sekwencji zdjęć dna oka wykonanych w długich, nieregularnych okresach. Znacząco przewyższa modele bazujące na pojedynczym obrazie, co silnie sugeruje korzyść z modelowania wzdłużnego dla prognozy chorób, gdzie uwzględnienie wcześniejszego obrazowania zwiększa wartość prognostyczną.15
  • Modele oparte na OCT siatkówki – analiza 2651 pomiarów OCT u 195 pacjentów z RRMS, 87 SPMS, 125 PPMS i 98 osób kontrolnych wykazała, że degeneracja siatkówki występowała i prognozowała aktywność choroby we wszystkich podtypach stwardnienia rozsianego. Grubość okołotarczowej i plamkowej warstwy włókien nerwowych siatkówki (pRNFL, mRNFL) przewidywała przyszłe rzuty we wszystkich podtypach MS i RRMS, podczas gdy grubość mRNFL i warstwy komórek zwojowych-wewnętrznej warstwy splotowatej (GCIPL) przewidywała przyszłą aktywność MRI w RRMS i PPMS.1617
  • Model łączący obrazowanie siatkówki z metadanymi klinicznymi – model, który integrował wyniki uzyskane z obrazów siatkówki i metadane kliniczne, wykazał znacznie lepszą wydajność predykcyjną w porównaniu do wykorzystania samych metadanych klinicznych w przewidywaniu chorób ogólnoustrojowych.18

Biomarkery prognostyczne w obrazowaniu siatkówki

Obrazowanie siatkówki dostarcza cennych biomarkerów prognostycznych dla różnych chorób:

  • Biomarkery naczyniowe – wyodrębniona za pomocą AI wazometria siatkówki (RV) oferuje alternatywny biomarker predykcyjny dla zdrowia naczyniowego w stosunku do tradycyjnych skal ryzyka, bez konieczności pobierania krwi czy pomiaru ciśnienia krwi. Modele predykcyjne dla śmiertelności z przyczyn krążeniowych u mężczyzn i kobiet miały skorygowane wartości C-statystyki i R² między 0,75-0,77 i 0,33-0,44.1920
  • Biomarkery w przewidywaniu chorób ogólnoustrojowych – sztuczna inteligencja wykorzystująca głębokie uczenie ma potencjał do przewidywania chorób ogólnoustrojowych na podstawie obrazowania siatkówki. Algorytmy DL wykazały skuteczność w identyfikacji cech obrazu siatkówki związanych z pogorszeniem funkcji poznawczych, demencją, chorobą Parkinsona i czynnikami ryzyka sercowo-naczyniowego.21
  • Biomarkery w IRD – postępy w technologii obrazowania siatkówki umożliwiły bardziej precyzyjną charakterystykę fenotypu IRD, co w konsekwencji prowadzi do lepszego zrozumienia patofizjologii tej grupy chorób. W ostatnich dziesięcioleciach odkryto szereg nowych biomarkerów obrazowych w kontekście IRD, które były wykorzystywane w badaniach klinicznych jako powtarzalne i precyzyjne miary wyników i punkty końcowe.22

Skuteczność prognozowania chorób siatkówki

Badania wykazują wysoką skuteczność modeli prognostycznych w różnych chorobach siatkówki:

  • AMD – dokładność prognostyczna podejść opartych na DL, mierzona za pomocą 5-letniej C-statystyki jako głównej miary wyniku, znacznie przewyższyła oba istniejące standardy. Ogólnie, analiza obrazu oparta na DL zapewniła dokładniejsze prognozy niż te pochodzące z oceny specjalistów siatkówki.23
  • Wielochorobowe modele diagnostyczne – wynik proponowanego modelu do wykrywania wielu chorób siatkówki mierzony w kategoriach dokładności walidacji i testów wynosił odpowiednio 89,81% i 88,72%.24 Badanie z wykorzystaniem konwolucyjnych sieci neuronowych wykazało najwyższą wydajność predykcyjną w przypadku zaćmy i patologicznej krótkowzroczności, uzyskując wynik F1 odpowiednio 96% i 94%.2526
  • Stwardnienie rozsiane – utrata 1 μm grubości pRNFL u pacjentów z RRMS bez historii zapalenia nerwu wzrokowego zwiększa prawdopodobieństwo nawrotu o 34%.27

Ograniczenia obecnych metod prognozowania

Pomimo obiecujących wyników, istnieją pewne ograniczenia w obecnych metodach prognozowania:

  • Zmienność pomiarów – przy obecnym stanie technologii, wzdłużne oceny grubości siatkówki mogą nie być odpowiednie na poziomie pojedynczego pacjenta. Wzdłużne zmiany grubości w rozsądnych odstępach czasu podlegały znacznym wahaniom pomiarowym i nie nadawały się do przewidywania aktywności choroby na poziomie pojedynczego pacjenta.28
  • Brak standaryzacji – dotychczas brakuje konsensusu co do optymalnych miar wyników i zastępczych punktów końcowych do zastosowania w badaniach klinicznych. Identyfikacja standaryzowanych, precyzyjnych i powtarzalnych zastępczych punktów końcowych jest ważna dla lepszego zrozumienia naturalnej historii tej niejednorodnej grupy chorób i poprawy oceny skuteczności terapeutycznej.29
  • Niesystematyczny przegląd literatury – ograniczenia badań obejmują przegląd literatury w sposób niesystematyczny, co może prowadzić do pominięcia niektórych artykułów lub nieodpowiedniego ustalenia priorytetów; oraz ograniczenia redakcyjne, które uniemożliwiły kompleksowy przegląd zastosowań AI we wszystkich zaburzeniach siatkówki.30

Przyszłość prognozowania chorób siatkówki

Przyszłość prognozowania chorób siatkówki wygląda obiecująco dzięki postępom w technologii i badaniach:

  • Personalizacja leczenia – dokładniejsze prognozowanie umożliwi bardziej spersonalizowane podejście do leczenia pacjentów z chorobami siatkówki, co potencjalnie poprawi wyniki leczenia.31
  • Szersza dostępność – badanie przesiewowe populacji ogólnej za pomocą nieinwazyjnych metod opartych na obrazowaniu siatkówki może być wykorzystane jako bezkontaktowa forma badania ogólnoustrojowego stanu zdrowia naczyniowego, aby selekcjonować osoby o średnim-wysokim ryzyku do dalszej klinicznej oceny ryzyka i odpowiedniej interwencji.32
  • Rola AI w przyszłości opieki zdrowotnej – AI reprezentuje jedno z ważnych podejść, aby sprostać wyzwaniom i ułatwić poprawę opieki nad pacjentami zarówno na poziomie indywidualnym, z bardziej terminowym, dokładnym i dostosowanym zarządzaniem, jak i na poziomie populacyjnym, w opiece zdrowotnej na dużą skalę. Przyszłość opieki zdrowotnej będzie w coraz większym stopniu uwzględniać korzyści, jakie może zapewnić AI, aby poprawić życie pacjentów i bez wątpienia wykonywać oceny szybciej i dokładniej niż specjaliści siatkówki mogą obecnie zrównoważenie zapewnić, pozwalając nam spędzać więcej czasu jako lepsi klinicyści i naukowcy.33

Znaczenie kliniczne prognozowania chorób siatkówki

Dokładne prognozowanie przebiegu chorób siatkówki ma istotne implikacje kliniczne:

  • Wczesna interwencja – wczesne wykrycie chorób siatkówki i właściwa diagnoza mogą zapobiec trwałej utracie wzroku. Osiągnięcia w dziedzinie terapii genowej i obrazowania siatkówki doprowadziły do niezwykłego postępu w projektowaniu badań klinicznych u ludzi i klinicznych testów nowych terapeutyków dla IRD.34
  • Udoskonalenie badań klinicznych – efektywne prowadzenie odpowiednio zasilanych badań klinicznych w retinitis pigmentosa jest utrudnione przez konieczność przesiewowego badania stosunkowo dużej liczby pacjentów, aby znaleźć tych, którzy pasują do kryteriów włączenia. Algorytmy DL mogą pomóc w efektywnym przesiewaniu potencjalnych uczestników w przyszłych badaniach lub próbach klinicznych RP.35
  • Wsparcie dla niespecjalistów – osiągnięcia w prognozowaniu chorób siatkówki mogą znacznie przyczynić się do jakości opieki medycznej poprzez ułatwienie wczesnej diagnozy, zwłaszcza przez niespecjalistów, dostępu do opieki, zmniejszenie kosztów skierowań i zapobieganie niepotrzebnym badaniom klinicznym i genetycznym.36 Biorąc pod uwagę, że okuliści są kluczowi dla diagnozy chorób siatkówki, a ich okres szkolenia wynosi 12 lat, istnieje dotkliwy niedobór okulistów w wielu społecznościach, co podkreśla wartość zautomatyzowanych systemów diagnostycznych.37

Podsumowując, prognozowanie chorób siatkówki przechodzi znaczącą transformację dzięki zastosowaniu zaawansowanych technik obrazowania i sztucznej inteligencji. Modele oparte na uczeniu głębokim osiągają coraz wyższą dokładność w przewidywaniu przebiegu chorób siatkówki, co może prowadzić do bardziej spersonalizowanego podejścia w leczeniu i monitorowaniu pacjentów. Pomimo pewnych ograniczeń, przyszłość prognozowania w tej dziedzinie wygląda obiecująco, oferując potencjał do znacznej poprawy wyników leczenia i jakości życia pacjentów z chorobami siatkówki.38

Kolejne rozdziały

Zapraszamy do dalszego czytania naszego leksykonu.

Wybierz kolejny rozdział z menu poniżej, aby otworzyć nową podstronę kompedium wiedzy i uzyskać szczegółowe informację o leku, substancji lub chorobie.

  1. 11.04.2026
  2. www.leksykon.com.pl

Materiały źródłowe

  • #1
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    Retinal diseases are a leading cause of blindness in developed countries, accounting for the largest share of visually impaired children, working-age adults (inherited retinal disease), and elderly individuals (age-related macular degeneration). […] For rare diseases, it can take several years for a final diagnosis (diagnostic odyssey), resulting in uncertainty about the prognosis and delay in appropriate care. […] The use of AI with DL tools has great potential in AMD, both for diagnostic purposes while allowing for a more efficient and accurate approach to prognostication of affected individuals and perhaps to directly determine (predict) efficacy of treatments. […] A more recent study attempted to introduce an AI system that combines 3D OCT images and automatic tissue maps in individuals with unilateral nAMD to predict progression in the contralateral eye.
  • #2 A deep learning framework for the early detection of multi-retinal diseases | PLOS One
    https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0307317
    Retinal diseases, such as DR, Age-Related Molecular Degeneration (ARMD), and glaucoma, are major contributors to blindness on a global scale. Timely identification and precise recovery from these conditions are essential for prompt treatment and the prevention of vision loss. […] Early detection of these diseases and proper diagnosis may prevent permanent vision loss. With appropriate treatment and consistent monitoring, it is feasible to decelerate or hinder additional deterioration of vision, particularly when the condition is identified during its initial phases. […] Early detection of DR is crucial for successful treatment and to avoid poor visual outcomes. […] The accuracy rates suggested that the deep learning model has learned to distinguish between different eye disease categories and healthy images effectively.
  • #3 Predicting risk of late age-related macular degeneration using deep learning
    https://pmc.ncbi.nlm.nih.gov/articles/PMC7453007/
    By 2040, age-related macular degeneration (AMD) will affect ~288 million people worldwide. Identifying individuals at high risk of progression to late AMD, the sight-threatening stage, is critical for clinical actions, including medical interventions and timely monitoring. […] When validated against an independent test data set of 601 participants, our model achieved high prognostic accuracy (5-year C-statistic 86.4 (95% confidence interval 86.286.6)) that substantially exceeded that of retinal specialists using two existing clinical standards (81.3 (81.181.5) and 82.0 (81.882.3), respectively). […] These results highlight the potential of deep learning systems to enhance clinical decision-making in AMD patients. […] Making accurate time-based predictions of progression to late AMD is clinically critical. This would enable improved decision-making regarding: (i) medical treatments, especially oral supplements known to decrease progression risk, (ii) lifestyle interventions, particularly smoking cessation and dietary changes, and (iii) intensity of patient monitoring, e.g., frequent reimaging in clinic and/or tailored home monitoring programs.
  • #4
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    Retinal diseases are a leading cause of blindness in developed countries, accounting for the largest share of visually impaired children, working-age adults (inherited retinal disease), and elderly individuals (age-related macular degeneration). […] For rare diseases, it can take several years for a final diagnosis (diagnostic odyssey), resulting in uncertainty about the prognosis and delay in appropriate care. […] The use of AI with DL tools has great potential in AMD, both for diagnostic purposes while allowing for a more efficient and accurate approach to prognostication of affected individuals and perhaps to directly determine (predict) efficacy of treatments. […] A more recent study attempted to introduce an AI system that combines 3D OCT images and automatic tissue maps in individuals with unilateral nAMD to predict progression in the contralateral eye.
  • #5 Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling | npj Digital Medicine
    https://www.nature.com/articles/s41746-024-01207-4
    Deep learning has enabled breakthroughs in automated diagnosis from medical imaging, with many successful applications in ophthalmology. […] For slow, progressive eye diseases like age-related macular degeneration (AMD) and primary open-angle glaucoma (POAG), patients undergo repeated imaging over time to track disease progression and forecasting the future risk of developing a disease is critical to properly plan treatment. Our proposed Longitudinal Transformer for Survival Analysis (LTSA) enables dynamic disease prognosis from longitudinal medical imaging, modeling the time to disease from sequences of fundus photography images captured over long, irregular time periods. […] A temporal attention analysis also suggested that, while the most recent image is typically the most influential, prior imaging still provides additional prognostic value.
  • #6 Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling | npj Digital Medicine
    https://www.nature.com/articles/s41746-024-01207-4
    Overall, these prior efforts often formulate automated prognosis as a binary classification task, for example, predicting whether a patient will develop the disease within fixed durations from the last visit (e.g., 2-year or 5-year prognosis). […] To avoid these pitfalls, we adopt a survival analysis approach to disease prognosis from longitudinal imaging data, aiming to model a time-to-event outcome (e.g., years until death or developing a disease) based on time-varying patient measurements. Such an approach is far more flexible and clinically valuable than prior efforts toward eye disease prognosis, as it incorporates longitudinal patient imaging and produces dynamic and long-term risk assessments. […] For both late AMD and POAG, accurately identifying high-risk patients as early as possible is critical to clinical decision-making, helping inform management, treatment planning, or patient monitoring.
  • #7 Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling | npj Digital Medicine
    https://www.nature.com/articles/s41746-024-01207-4
    Deep learning has enabled breakthroughs in automated diagnosis from medical imaging, with many successful applications in ophthalmology. […] For slow, progressive eye diseases like age-related macular degeneration (AMD) and primary open-angle glaucoma (POAG), patients undergo repeated imaging over time to track disease progression and forecasting the future risk of developing a disease is critical to properly plan treatment. Our proposed Longitudinal Transformer for Survival Analysis (LTSA) enables dynamic disease prognosis from longitudinal medical imaging, modeling the time to disease from sequences of fundus photography images captured over long, irregular time periods. […] A temporal attention analysis also suggested that, while the most recent image is typically the most influential, prior imaging still provides additional prognostic value.
  • #8
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    The age-related eye disease studies (AREDS and AREDS2) used DL algorithms and survival analysis to predict risk of late AMD, which achieved high prognostic accuracy. […] AI algorithms using multimodal imaging techniques have been developed to facilitate the diagnosis, classification, decipher the genetic aetiology, and measure the progression rate of IRD. […] Predicting VA based on OCT and infrared images in RP has been assessed by Liu et al. They were able to determine if a patient with RP had VA below or above 20/40, with an AUC of 0.85. […] ROP is an important cause of preventable childhood blindness worldwide. […] These study limitations are the review of the literature in a non-systematic approach, possibly leading to some papers being omitted or not adequately prioritised; and editorial restrictions, which prevented us from doing a comprehensive review of AI applications in all retinal disorders.
  • #9 Predicting risk of late age-related macular degeneration using deep learning
    https://pmc.ncbi.nlm.nih.gov/articles/PMC7453007/
    By 2040, age-related macular degeneration (AMD) will affect ~288 million people worldwide. Identifying individuals at high risk of progression to late AMD, the sight-threatening stage, is critical for clinical actions, including medical interventions and timely monitoring. […] When validated against an independent test data set of 601 participants, our model achieved high prognostic accuracy (5-year C-statistic 86.4 (95% confidence interval 86.286.6)) that substantially exceeded that of retinal specialists using two existing clinical standards (81.3 (81.181.5) and 82.0 (81.882.3), respectively). […] These results highlight the potential of deep learning systems to enhance clinical decision-making in AMD patients. […] Making accurate time-based predictions of progression to late AMD is clinically critical. This would enable improved decision-making regarding: (i) medical treatments, especially oral supplements known to decrease progression risk, (ii) lifestyle interventions, particularly smoking cessation and dietary changes, and (iii) intensity of patient monitoring, e.g., frequent reimaging in clinic and/or tailored home monitoring programs.
  • #10 Predicting risk of late age-related macular degeneration using deep learning
    https://pmc.ncbi.nlm.nih.gov/articles/PMC7453007/
    The prediction accuracy of the approaches was compared using the 5-year C-statistic as the primary outcome measure. The 5-year C-statistic of the two DL approaches met and substantially exceeded that of both existing standards. […] Overall, DL-based image analysis provided more accurate predictions than those from retinal specialist grading using the two existing standards. […] In conclusion, combining DL feature extraction of CFP with survival analysis achieved high prognostic accuracy in predictions of progression to late AMD, and its subtypes, over a wide time interval (112 years). Not only did its accuracy meet and surpass existing clinical standards, but additional strengths in clinical settings include risk ascertainment above 50% and without genotype data.
  • #11
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    The age-related eye disease studies (AREDS and AREDS2) used DL algorithms and survival analysis to predict risk of late AMD, which achieved high prognostic accuracy. […] AI algorithms using multimodal imaging techniques have been developed to facilitate the diagnosis, classification, decipher the genetic aetiology, and measure the progression rate of IRD. […] Predicting VA based on OCT and infrared images in RP has been assessed by Liu et al. They were able to determine if a patient with RP had VA below or above 20/40, with an AUC of 0.85. […] ROP is an important cause of preventable childhood blindness worldwide. […] These study limitations are the review of the literature in a non-systematic approach, possibly leading to some papers being omitted or not adequately prioritised; and editorial restrictions, which prevented us from doing a comprehensive review of AI applications in all retinal disorders.
  • #12 Prediction of causative genes in inherited retinal disorder from fundus photography and autofluorescence imaging using deep learning techniques – PubMed
    https://pubmed.ncbi.nlm.nih.gov/33879469/
    To investigate the utility of a data-driven deep learning approach in patients with inherited retinal disorder (IRD) and to predict the causative genes based on fundus photography and fundus autofluorescence (FAF) imaging. […] A novel application of deep neural networks in the prediction of the causative IRD genes from fundus photographs and FAF, with a high prediction accuracy of over 80%, was highlighted. These achievements will extensively promote the quality of medical care by facilitating early diagnosis, especially by non-specialists, access to care, reducing the cost of referrals, and preventing unnecessary clinical and genetic testing.
  • #13
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    The age-related eye disease studies (AREDS and AREDS2) used DL algorithms and survival analysis to predict risk of late AMD, which achieved high prognostic accuracy. […] AI algorithms using multimodal imaging techniques have been developed to facilitate the diagnosis, classification, decipher the genetic aetiology, and measure the progression rate of IRD. […] Predicting VA based on OCT and infrared images in RP has been assessed by Liu et al. They were able to determine if a patient with RP had VA below or above 20/40, with an AUC of 0.85. […] ROP is an important cause of preventable childhood blindness worldwide. […] These study limitations are the review of the literature in a non-systematic approach, possibly leading to some papers being omitted or not adequately prioritised; and editorial restrictions, which prevented us from doing a comprehensive review of AI applications in all retinal disorders.
  • #14 Prediction of visual impairment in retinitis pigmentosa using deep learning and multimodal fundus images | British Journal of Ophthalmology
    https://bjo.bmj.com/content/107/10/1484
    The efficient conduct of adequately powered clinical trials in retinitis pigmentosa (RP) is hampered by the need to screen relatively large numbers of patients to find those that fit the inclusion criteria. […] Structure-function correlation based solely on confocal scanning laser ophthalmoscopy imaging in patients with RP can be predicted using deep learning (DL). […] DL-based estimation of visual acuity using optical coherence tomography images may enable efficient screening of potential subjects in future RP research studies or clinical trials. […] Our algorithm showed robust performance in predicting visual impairment in patients with RP, thus providing proof-of-concept for predicting structure-function correlation based solely on cSLO imaging in patients with RP. […] A DL algorithm can discriminate between two levels of VA with relatively high sensitivity and specificity, using only a single-slice transfoveal OCT image as the input data. Specifically, the DL algorithm was able to detect visual impairment based on a VA cut-off of 20/40. The role of multimodal imaging input in improving algorithm performance is unclear at present. These data establish the feasibility of predicting structure-function correlation based on OCT images in patients with RP.
  • #15 Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling | npj Digital Medicine
    https://www.nature.com/articles/s41746-024-01207-4
    Our proposed LTSA is trained on sequences of consecutive fundus images to directly predict the time-varying hazard distribution, allowing for disease prognosis through a survival analysis framework. […] LTSA significantly outperformed the single-image baseline on 37 out of 40 head-to-head comparisons across a wide variety of prediction, or landmark times, and time horizons. […] Our results also strongly suggest the benefit of longitudinal image modeling for prognosis, where incorporating prior imaging enhances disease prognosis. […] This study offers a potential answer to the growing demand for dynamic and explainable prognoses for eye diseases. […] Our results suggest that longitudinal modeling can improve eye disease risk prognosis, providing evidence that prior imaging can provide added prognostic value.
  • #16 Evolution of retinal degeneration and prediction of disease activity in relapsing and progressive multiple sclerosis | Nature Communications
    https://www.nature.com/articles/s41467-024-49309-7
    Retinal optical coherence tomography has been identified as biomarker for disease progression in relapsing-remitting multiple sclerosis (RRMS), while the dynamics of retinal atrophy in progressive MS are less clear. […] Here, we analyzed 2651 OCT measurements of 195 RRMS, 87 SPMS, 125 PPMS patients, and 98 controls from five German MS centers after quality control. […] Peripapillary and macular retinal nerve fiber layer (pRNFL, mRNFL) thickness predicted future relapses in all MS and RRMS patients while mRNFL and ganglion cell-inner plexiform layer (GCIPL) thickness predicted future MRI activity in RRMS (mRNFL, GCIPL) and PPMS (GCIPL). […] mRNFL thickness predicted future disability progression in PPMS. […] In conclusion, retinal degeneration, most pronounced of pRNFL and GCIPL, occurs in all subtypes.
  • #17 Evolution of retinal degeneration and prediction of disease activity in relapsing and progressive multiple sclerosis | Nature Communications
    https://www.nature.com/articles/s41467-024-49309-7
    Using the current state of technology, longitudinal assessments of retinal thickness may not be suitable on a single patient level. […] Analyzing 2651 OCT measurements of 195 RRMS, 87 SPMS, 125 PPMS patients and 98 controls from five German MS centers, we could demonstrate that retinal degeneration occurred and predicted disease activity in all MS subtypes. […] However, longitudinal thickness change rates over reasonable intervals were subject to considerable amounts of measurement variability and not suitable to predict disease activity on a single patient level. […] When adjusting for all covariates in the LMER, lower pRNFL and mRNFL thickness were associated with increased probability for relapses in all MS and RRMS patients without ON (pRNFL and mRNFL) and all RRMS patients (pRNFL).
  • #18 Prognostic potentials of AI in ophthalmology: systemic disease forecasting via retinal imaging | Eye and Vision | Full Text
    https://eandv.biomedcentral.com/articles/10.1186/s40662-024-00384-3
    Here, we aim to explore the traditional value of the retina for systemic disease assessment, examine the potential of AI-based retinal biomarkers in predicting various systemic diseases, and emphasize the importance of longitudinal prediction models for early detection and personalized care. […] AI-based retinal biomarkers have emerged as a promising approach for the early detection and monitoring of neurodegenerative diseases. […] The findings revealed that the combined model, which integrated risk scores extracted from retinal images and clinical metadata, demonstrated significantly enhanced predictive performance compared to utilizing clinical metadata alone. […] This indicates the potential of retinal images as a valuable screening tool for risk assessment and personalized treatment in the context of CKD. […] Overall, the study showcases the potential of an AI-based biomarker, the Reti-CKD score, in a non-invasive way for predicting the risk of CKD development by leveraging DL algorithms trained on retinal photographs and incorporating clinical factors.
  • #19 Artificial intelligence-enabled retinal vasculometry for prediction of circulatory mortality, myocardial infarction and stroke | British Journal of Ophthalmology
    https://bjo.bmj.com/content/106/12/1722
    Aims We examine whether inclusion of artificial intelligence (AI)-enabled retinal vasculometry (RV) improves existing risk algorithms for incident stroke, myocardial infarction (MI) and circulatory mortality. […] Prediction models for circulatory mortality in men and women had optimism adjusted C-statistics and R2 statistics between 0.750.77 and 0.330.44, respectively. […] RV offers an alternative predictive biomarker to traditional risk-scores for vascular health, without the need for blood sampling or blood pressure measurement. Further work is needed to examine RV in population screening to triage individuals at high-risk. […] Risk models developed in UK Biobank (validated in European Prospective Investigation into Cancer-Norfolk) using artificial intelligence (AI)-enabled retinal vasculometry (RV), age, history of cardiovascular disease, use of hypertensive medication and smoking yielded high predictive test performance for circulatory mortality.
  • #20 Artificial intelligence-enabled retinal vasculometry for prediction of circulatory mortality, myocardial infarction and stroke | British Journal of Ophthalmology
    https://bjo.bmj.com/content/106/12/1722
    AI-enabled RV extraction offers a non-invasive prognostic biomarker of vascular health that does not require blood sampling or blood pressure measurement, and potentially has greater community reach to identify individuals at medium-high risk requiring further clinical assessment. […] Our automated AI-enabled system extracts the retinal vascular tree over the entire retinal image, distinguishes between arterioles and venules and provides measures of tortuosity, width-variance and area, in addition to vessel width. […] Prediction of circulatory mortality using age, sex, smoking status, medical history and RV has not been reported previously, and yielded the highest model performance in terms of C-statistics R2 statistics and agreement between observed and predicted risks, even at lower levels of risk, in both the internal and external validation cohorts.
  • #21 Prognostic potentials of AI in ophthalmology: systemic disease forecasting via retinal imaging | Eye and Vision | Full Text
    https://eandv.biomedcentral.com/articles/10.1186/s40662-024-00384-3
    Artificial intelligence (AI) that utilizes deep learning (DL) has potential for systemic disease prediction using retinal imaging. […] This review highlights the substantial potential of AI-based retinal biomarkers in predicting neurodegenerative, cardiovascular, and chronic kidney diseases. Notably, DL algorithms have demonstrated effectiveness in identifying retinal image features associated with cognitive decline, dementia, Parkinson’s disease, and cardiovascular risk factors. […] AI-based retinal imaging hold promise in transforming primary care and systemic disease management. […] Utilizing baseline retinal photos, longitudinal prediction models forecast the likelihood of future systemic diseases such as CVD and chronic kidney disease (CKD). […] AI has demonstrated significant promise in quantifiable risk assessment in specific contexts, where DL models have been rigorously compared to human assessment, indicating its potential to enhance disease prediction and management strategies.
  • #22 Retinal Imaging Findings in Inherited Retinal Diseases
    https://www.mdpi.com/2077-0383/13/7/2079
    The aim of this narrative review is to describe the advances in retinal imaging for patients with inherited retinal diseases, their characteristic imaging findings, and the imaging endpoint measures used in clinical research to serve as a guide for clinicians and trainees working in a clinical setting. […] Advances in retinal imaging technology have allowed a more precise characterization of the phenotype of IRDs, and consequently an improvement in the understanding of the pathophysiology of this group of diseases. […] Over the last decades, a number of new imaging biomarkers in the context of IRDs have been uncovered and employed in clinical trials as repeatable and precise outcome measures and endpoints, facilitating clinical research and therapeutic development for IRDs. […] Ultimately, we expect these advances to allow us to more precisely diagnose and monitor disease, as well as the response to therapy, which will better care for our patients with IRD.
  • #23 Predicting risk of late age-related macular degeneration using deep learning
    https://pmc.ncbi.nlm.nih.gov/articles/PMC7453007/
    The prediction accuracy of the approaches was compared using the 5-year C-statistic as the primary outcome measure. The 5-year C-statistic of the two DL approaches met and substantially exceeded that of both existing standards. […] Overall, DL-based image analysis provided more accurate predictions than those from retinal specialist grading using the two existing standards. […] In conclusion, combining DL feature extraction of CFP with survival analysis achieved high prognostic accuracy in predictions of progression to late AMD, and its subtypes, over a wide time interval (112 years). Not only did its accuracy meet and surpass existing clinical standards, but additional strengths in clinical settings include risk ascertainment above 50% and without genotype data.
  • #24 A deep learning framework for the early detection of multi-retinal diseases | PLOS One
    https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0307317
    The outcome of the suggested model is measured in terms of validation and testing accuracies, which are 89.81% and 88.72%, respectively. […] This study incorporates a comparative analysis of prior research, with a specific emphasis on the RFMiD dataset. Compared to previous studies, our research provides a more accurate method of detecting retinal diseases.
  • #25 Automated diagnosis of 7 retinal diseases with convolutional neural networks in a dataset of 2,234 eye images | Young Scientist Journal | Vanderbilt University
    https://wp0.vanderbilt.edu/youngscientistjournal/article/automated-diagnosis-of-7-retinal-diseases-with-convolutional-neural-networks-in-a-dataset-of-2234-eye-images
    Ophthalmologists are critical for diagnosis of retinal diseases, however with a training period of 12 years there is an acute shortage of ophthalmologists in many communities. […] In this study, we utilize a dataset including eight retinal disease outcomes and train five residual network models corresponding to each of the disease outcomes in the training set. […] We find the highest predictive performance with cataract and pathological myopia obtaining an F1 score of 96% and 94% respectively. […] Our study showed that convolutional neural networks most confidently diagnose cataracts and pathological myopia with testing F1 scores 96% and 94% respectively. […] Transferring knowledge by pre-training models on a related but distinct outcome lead to further improvements in performance from 94% to 96% testing F1 score in the diagnosis of cataract, without collecting further data.
  • #26 Automated diagnosis of 7 retinal diseases with convolutional neural networks in a dataset of 2,234 eye images | Young Scientist Journal | Vanderbilt University
    https://wp0.vanderbilt.edu/youngscientistjournal/article/automated-diagnosis-of-7-retinal-diseases-with-convolutional-neural-networks-in-a-dataset-of-2234-eye-images
    Diagnosis of pathological myopia and cataract simultaneously resulted in F1 scores of 94% and 86%, respectively. […] The findings from our experiments can help with two main things. First off, the findings from the experiment have helped leverage the fact that cataract and pathological myopia are two of the diseases best diagnosed by machine learning. […] By figuring this out, it can lead to future research being done on what makes these models better than others in diagnosing retinal diseases and can also lead to a better F1 score and accuracy when it comes to future endeavors.
  • #27 Evolution of retinal degeneration and prediction of disease activity in relapsing and progressive multiple sclerosis | Nature Communications
    https://www.nature.com/articles/s41467-024-49309-7
    mRNFL and GCIPL thickness predicted future MRI progression/activity in RRMS without ON (mRNFL and GCIPL thickness) and PPMS patients (GCIPL thickness). […] When adjusting for age, sex, and DMT in the LMER, mRNFL thickness predicted future disability progression in PPMS. […] The significance levels and odds ratios are provided in Tables 24. […] As an example, 1m of pRNFL thickness loss in RRMS patients without history of ON increases the likelihood of relapse by 34%. […] Interestingly, lower GCIPL thickness (Table 2) and mRNFL, GCIPL, and INL thickness of the lowest tertile (mRNFL29m, GCIPL 62m, INL33m) were associated with increased risk for future MRI progression/activity in PPMS patients.
  • #28 Evolution of retinal degeneration and prediction of disease activity in relapsing and progressive multiple sclerosis | Nature Communications
    https://www.nature.com/articles/s41467-024-49309-7
    Using the current state of technology, longitudinal assessments of retinal thickness may not be suitable on a single patient level. […] Analyzing 2651 OCT measurements of 195 RRMS, 87 SPMS, 125 PPMS patients and 98 controls from five German MS centers, we could demonstrate that retinal degeneration occurred and predicted disease activity in all MS subtypes. […] However, longitudinal thickness change rates over reasonable intervals were subject to considerable amounts of measurement variability and not suitable to predict disease activity on a single patient level. […] When adjusting for all covariates in the LMER, lower pRNFL and mRNFL thickness were associated with increased probability for relapses in all MS and RRMS patients without ON (pRNFL and mRNFL) and all RRMS patients (pRNFL).
  • #29 Retinal Imaging Findings in Inherited Retinal Diseases
    https://www.mdpi.com/2077-0383/13/7/2079
    Inherited retinal diseases (IRDs) represent one of the major causes of progressive and irreversible vision loss in the working-age population. […] Advances in gene therapy and retinal imaging have driven remarkable progress in the design of human clinical trials and clinical testing of novel therapeutics for IRDs. […] However, to date, there is a lack of consensus on the optimal outcome measures and surrogate endpoints for use in clinical trials. […] The identification of standardized, precise, and reproducible surrogate endpoints is important to achieve a better understanding of the natural history of this heterogeneous group of diseases and to improve the assessment of therapeutic efficacy. […] Consequently, there has been a growing interest in leveraging retinal imaging to uncover imaging endpoints, which could possibly be investigated as surrogate outcome measures in clinical trials.
  • #30
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    The age-related eye disease studies (AREDS and AREDS2) used DL algorithms and survival analysis to predict risk of late AMD, which achieved high prognostic accuracy. […] AI algorithms using multimodal imaging techniques have been developed to facilitate the diagnosis, classification, decipher the genetic aetiology, and measure the progression rate of IRD. […] Predicting VA based on OCT and infrared images in RP has been assessed by Liu et al. They were able to determine if a patient with RP had VA below or above 20/40, with an AUC of 0.85. […] ROP is an important cause of preventable childhood blindness worldwide. […] These study limitations are the review of the literature in a non-systematic approach, possibly leading to some papers being omitted or not adequately prioritised; and editorial restrictions, which prevented us from doing a comprehensive review of AI applications in all retinal disorders.
  • #31 Predicting risk of late age-related macular degeneration using deep learning
    https://pmc.ncbi.nlm.nih.gov/articles/PMC7453007/
    By 2040, age-related macular degeneration (AMD) will affect ~288 million people worldwide. Identifying individuals at high risk of progression to late AMD, the sight-threatening stage, is critical for clinical actions, including medical interventions and timely monitoring. […] When validated against an independent test data set of 601 participants, our model achieved high prognostic accuracy (5-year C-statistic 86.4 (95% confidence interval 86.286.6)) that substantially exceeded that of retinal specialists using two existing clinical standards (81.3 (81.181.5) and 82.0 (81.882.3), respectively). […] These results highlight the potential of deep learning systems to enhance clinical decision-making in AMD patients. […] Making accurate time-based predictions of progression to late AMD is clinically critical. This would enable improved decision-making regarding: (i) medical treatments, especially oral supplements known to decrease progression risk, (ii) lifestyle interventions, particularly smoking cessation and dietary changes, and (iii) intensity of patient monitoring, e.g., frequent reimaging in clinic and/or tailored home monitoring programs.
  • #32 Artificial intelligence-enabled retinal vasculometry for prediction of circulatory mortality, myocardial infarction and stroke | British Journal of Ophthalmology
    https://bjo.bmj.com/content/106/12/1722
    Our approach of focusing on the retinal microvasculature as a key prognostic marker of incident cardiovascular outcomes and circulatory mortality is supported by saliency maps presented in a study using end-to-end AI of retinal images to estimate the extent of coronary artery calcium scores in cross-sectional associations, with C-statistics for incident CVD varying between 0.68 and 0.76. […] In the general population it could be used as a non-contact form of systemic vascular health check, to triage those at medium-high risk of circulatory mortality for further clinical risk assessment and appropriate intervention.
  • #33
    https://link.springer.com/article/10.1007/s00417-023-06052-x
    AI represents one important approach to help meet these challenges and moreover facilitate improvements in patient care both at the individual level with more timely, accurate, and bespoke management, as well as population-level, large-scale healthcare. […] The future of healthcare will increasingly incorporate the advantages that AI can provide to improve the lives of our patients and no doubt perform assessments quicker and more accurately than retina specialists can currently sustainably provide, allowing us to spend more time being better clinicians and scientists.
  • #34 Retinal Imaging Findings in Inherited Retinal Diseases
    https://www.mdpi.com/2077-0383/13/7/2079
    Inherited retinal diseases (IRDs) represent one of the major causes of progressive and irreversible vision loss in the working-age population. […] Advances in gene therapy and retinal imaging have driven remarkable progress in the design of human clinical trials and clinical testing of novel therapeutics for IRDs. […] However, to date, there is a lack of consensus on the optimal outcome measures and surrogate endpoints for use in clinical trials. […] The identification of standardized, precise, and reproducible surrogate endpoints is important to achieve a better understanding of the natural history of this heterogeneous group of diseases and to improve the assessment of therapeutic efficacy. […] Consequently, there has been a growing interest in leveraging retinal imaging to uncover imaging endpoints, which could possibly be investigated as surrogate outcome measures in clinical trials.
  • #35 Prediction of visual impairment in retinitis pigmentosa using deep learning and multimodal fundus images | British Journal of Ophthalmology
    https://bjo.bmj.com/content/107/10/1484
    The efficient conduct of adequately powered clinical trials in retinitis pigmentosa (RP) is hampered by the need to screen relatively large numbers of patients to find those that fit the inclusion criteria. […] Structure-function correlation based solely on confocal scanning laser ophthalmoscopy imaging in patients with RP can be predicted using deep learning (DL). […] DL-based estimation of visual acuity using optical coherence tomography images may enable efficient screening of potential subjects in future RP research studies or clinical trials. […] Our algorithm showed robust performance in predicting visual impairment in patients with RP, thus providing proof-of-concept for predicting structure-function correlation based solely on cSLO imaging in patients with RP. […] A DL algorithm can discriminate between two levels of VA with relatively high sensitivity and specificity, using only a single-slice transfoveal OCT image as the input data. Specifically, the DL algorithm was able to detect visual impairment based on a VA cut-off of 20/40. The role of multimodal imaging input in improving algorithm performance is unclear at present. These data establish the feasibility of predicting structure-function correlation based on OCT images in patients with RP.
  • #36 Prediction of causative genes in inherited retinal disorder from fundus photography and autofluorescence imaging using deep learning techniques – PubMed
    https://pubmed.ncbi.nlm.nih.gov/33879469/
    To investigate the utility of a data-driven deep learning approach in patients with inherited retinal disorder (IRD) and to predict the causative genes based on fundus photography and fundus autofluorescence (FAF) imaging. […] A novel application of deep neural networks in the prediction of the causative IRD genes from fundus photographs and FAF, with a high prediction accuracy of over 80%, was highlighted. These achievements will extensively promote the quality of medical care by facilitating early diagnosis, especially by non-specialists, access to care, reducing the cost of referrals, and preventing unnecessary clinical and genetic testing.
  • #37 Automated diagnosis of 7 retinal diseases with convolutional neural networks in a dataset of 2,234 eye images | Young Scientist Journal | Vanderbilt University
    https://wp0.vanderbilt.edu/youngscientistjournal/article/automated-diagnosis-of-7-retinal-diseases-with-convolutional-neural-networks-in-a-dataset-of-2234-eye-images
    Ophthalmologists are critical for diagnosis of retinal diseases, however with a training period of 12 years there is an acute shortage of ophthalmologists in many communities. […] In this study, we utilize a dataset including eight retinal disease outcomes and train five residual network models corresponding to each of the disease outcomes in the training set. […] We find the highest predictive performance with cataract and pathological myopia obtaining an F1 score of 96% and 94% respectively. […] Our study showed that convolutional neural networks most confidently diagnose cataracts and pathological myopia with testing F1 scores 96% and 94% respectively. […] Transferring knowledge by pre-training models on a related but distinct outcome lead to further improvements in performance from 94% to 96% testing F1 score in the diagnosis of cataract, without collecting further data.
  • #38 Retinal Imaging Findings in Inherited Retinal Diseases
    https://www.mdpi.com/2077-0383/13/7/2079
    The aim of this narrative review is to describe the advances in retinal imaging for patients with inherited retinal diseases, their characteristic imaging findings, and the imaging endpoint measures used in clinical research to serve as a guide for clinicians and trainees working in a clinical setting. […] Advances in retinal imaging technology have allowed a more precise characterization of the phenotype of IRDs, and consequently an improvement in the understanding of the pathophysiology of this group of diseases. […] Over the last decades, a number of new imaging biomarkers in the context of IRDs have been uncovered and employed in clinical trials as repeatable and precise outcome measures and endpoints, facilitating clinical research and therapeutic development for IRDs. […] Ultimately, we expect these advances to allow us to more precisely diagnose and monitor disease, as well as the response to therapy, which will better care for our patients with IRD.

Zobacz więcej

zaćma terapia genowa autofluorescencja dna oka zapalenie nerwu wzrokowego choroba siatkówki warstwa włókien nerwowych siatkówki zwyrodnienie plamki żółtej związane z wiekiem czynnik ryzyka sercowo-naczyniowego sztuczna inteligencja obrazowanie siatkówki retinitis pigmentosa sieć neuronowa jaskra pierwotna otwartego kąta trwała utrata wzroku funkcja poznawcza stwardnienie rozsiane choroba Parkinsona demencja patologiczna krótkowzroczność dziedziczna choroba siatkówki biomarker prognostyczny uczenie głębokie obrazowanie OCT upośledzenie widzenia utrata wzroku krótkowzroczność