RESUMO
OBJECTIVE: Artificial intelligence (AI) holds the potential to make significant advancements in pathology. However, its actual implementation and certification for practical use are currently limited, often due to challenges related to model transferability. In this context, we investigate the factors influencing transferability and present methods aimed at enhancing the utilization of AI algorithms in pathology. MATERIALS AND METHODS: Various convolutional neural networks (CNNs) and vision transformers (ViTs) were trained using datasets from two institutions, along with the publicly available TCGA-MIBC dataset. These networks conducted predictions in urothelial tissue and intrahepatic cholangiocarcinoma (iCCA). The objective was to illustrate the impact of stain normalization, the influence of various artifacts during both training and testing, as well as the effects of the NoisyEnsemble method. RESULTS: We were able to demonstrate that stain normalization of slides from different institutions has a significant positive effect on the inter-institutional transferability of CNNs and ViTs (respectively +13% and +10%). In addition, ViTs usually achieve a higher accuracy in the external test (here +1.5%). Similarly, we showcased how artifacts in test data can negatively affect CNN predictions and how incorporating these artifacts during training leads to improvements. Lastly, NoisyEnsembles of CNNs (better than ViTs) were shown to enhance transferability across different tissues and research questions (+7% Bladder, +15% iCCA). DISCUSSION: It is crucial to be aware of the transferability challenge: achieving good performance during development does not necessarily translate to good performance in real-world applications. The inclusion of existing methods to enhance transferability, such as stain normalization and NoisyEnsemble, and their ongoing refinement, is of importance.
Assuntos
Inteligência Artificial , Redes Neurais de Computação , Algoritmos , ArtefatosRESUMO
BACKGROUND: Metabolic dysfunction-associated steatotic liver disease (MASLD), or non-alcoholic fatty liver disease (NAFLD), is a common disease that is diagnosed through manual evaluation of liver biopsies, an assessment that is subject to high interobserver variability (IBV). IBV can be reduced using automated methods. OBJECTIVES: Many existing computer-based methods do not accurately reflect what pathologists evaluate in practice. The goal is to demonstrate how these differences impact the prediction of hepatic steatosis. Additionally, IBV complicates algorithm validation. MATERIALS AND METHODS: Forty tissue sections were analyzed to detect steatosis, nuclei, and fibrosis. Data generated from automated image processing were used to predict steatosis grades. To investigate IBV, 18 liver biopsies were evaluated by multiple observers. RESULTS: Area-based approaches yielded more strongly correlated results than nucleus-based methods (â Spearman rho [ρ]â¯= 0.92 vs. 0.79). The inclusion of information regarding tissue composition reduced the average absolute error for both area- and nucleus-based predictions by 0.5% and 2.2%, respectively. Our final area-based algorithm, incorporating tissue structure information, achieved a high accuracy (80%) and strong correlation (â Spearman ρâ¯= 0.94) with manual evaluation. CONCLUSION: The automatic and deterministic evaluation of steatosis can be improved by integrating information about tissue composition and can serve to reduce the influence of IBV.
Assuntos
Hepatopatia Gordurosa não Alcoólica , Humanos , Hepatopatia Gordurosa não Alcoólica/diagnóstico , Biópsia , Fibrose , AutomaçãoRESUMO
BACKGROUND: Fluorescence-based confocal microscopy (FCM) can be used to create virtual H&E sections in real time. So far, FCM has been used in dermato-, uro-, and gynecopathology. FCM allows the creation of a completely digitized frozen section, which could potentially replace conventional frozen sections in the future. OBJECTIVE: The aim of the current work is to implement FCM technology as a component of fully digitized processes in the pathological workflow. For this purpose, the current use of FCM in liver transplant pathology will be extended to other disciplines such as urology and otorhinolaryngology. MATERIALS AND METHODS: The FCM technique continues to be used prospectively on native tissue samples from potential donor livers. Conventional frozen sections are used comparatively to virtual FCM scans. RESULTS: The data show a nearly perfect agreement for the detection of cholangitis, fibrosis, and malignancy, and a high level of agreement for, e.g., macrovesicular steatosis, inflammation, steatohepatitis, and necrosis between virtual FCM scans and conventional routine diagnostic frozen sections. CONCLUSION: Since the availability of time- and cost-intensive frozen section diagnostics in the context of transplant pathology in continuous operation (24/7) is now only established at very few university centers in Germany due to an increasing shortage of specialists, the use of FCM could be an important building block in the current process leading towards a fully digitized pathology workflow and should thus be extended to various disciplines.