Um método semiautomático para estimação de fibrose intersticial

Abstract

The estimation of the fibrosis content in the interstitium, suggested that it may accompany immunological and infectious complications, which may be associated with a process of rejection of a transplanted kidney, is possible with the use of the Banff Classification, analyzing slides of renal biopsies stained with Dye Masson's Trichrome. The evaluation of Interstitial Fibrosis is semi-quantitative and causes subjectivity, which can be excessive, since the diagnosis and correct analysis of images is a complex task and highly dependent on the experience of the pathologists who perform it. In this scenario of reflection regarding the analysis of images of kidney biopsies, automatic methods to aid diagnosis using computational techniques can contribute to an increase in the quality of medical diagnosis. Considering that the Banff Classification is a relevant criterion for evaluating patients with chronic kidney disease modified by kidney transplantation, it is important to find ways to help mitigate subjectivity and intraand inter-examiner variability in the evaluation of indicators used in this classification, especially in the inherited degree of interstitial fibrosis. In this context, this work presents a computational method to search semi automatically for interstitial fibrosis, from digital images of renal biopsies, from the perspective of the Banff Classification. A dataset of 309 images was annotated by 5 pathologists and the results compared with the measurements made by the developed system. The results showed little agreement in the evaluation between the physicians, in addition to a large variance when they performed two consecutive measurements of the same image set. In relation to the quality of the system, it carried out the measurement invariably, but its precision could not be definitively evaluated, due to the variability obtained in the analysis of the pathologists. Due to the variable degree of reproducibility and agreement presented by the pathologists, it was not possible to assess the quality of the computational method, since a gold standard was not found in the medical team. Even so, this project built a semi-automatic model for evaluating interstitial fibrosis that is available as part of the functionality of the PathoSpotter project (https://pathospotter.bahia.fiocruz.br). The system can be used as an estimator of the degree of interstitial fibrosis in biopsy slide sections that are selected by the pathologist, and can be adjusted in the future by obtaining a dataset with consistent notes of interstitial fibrosis.