黑料网

A variety of systemic and corneal disorders may be linked to anomalies in corneal endothelial cells. By changing the hydration of the corneal stromal, endothelial cell damage can drastically reduce corneal transparency and cause irreversible endothelial cell pathology that necessitates corneal transplantation. Ophthalmologists have up to now undertaken quantitative examination of endothelial cell abnormalities manually utilizing time-consuming and extremely subjective semi-automatic technologies, which call for operator interaction. For the segmentation and computation of endothelial cells in pictures of the human cornea collected by in vivo corneal confocal microscopy, we created and implemented the Corneal Endothelium Analysis System (CEAS). To make the cells more visible, a Fast Fourier Transform (FFT) Band-pass filter is first used to minimize noise and improve the image quality. Second, to precisely measure the morphological properties of the human corneal endothelial cells, endothelial cell borders are found using watershed transformations and Voronoi tessellations.

Based on a library of 40 corneal confocal endothelial cell images, the performance of the automated segmentation system was compared to manually traced ground truth images in terms of segmentation accuracy and acquired clinical characteristics. Also, a second database of 40 pictures from controls (n=11), obese persons (n=16), and diabetic patients (n=13) was used to compare the resilience and effectiveness of the proposed CEAS system with manually determined cell densities. A Bland-Altman plot reveals that 95% of the data are between the 2SD agreement lines and that there is a Pearson correlation coefficient of 0.9 (p=0.0001) between the automated and manual endothelial cell densities. With an execution time of just 6 seconds per image, we show the efficiency and reliability of the CEAS system as well as the potential for its use in a real-world clinical scenario to facilitate speedy diagnosis and patient follow-up.

The cornerstone of cancer treatment is pathology. While individualized cancer therapy necessitates accurate biomarker assessment, the requirement for precision in histopathological cancer diagnosis is growing. The advent of digital image analysis offers hope for increasing the scope and accuracy of histomorphological assessment. Recent improvements in computational pathology have been made possible by machine learning, and particularly deep learning.

In the coming decade, the adoption of machine learning in ordinary healthcare will mark a turning point for the industry, and histopathology is at the epicentre of this change. Model-based evaluation of common diagnostic features in pathology and the capacity to extract and find novel features that offer insights into a disease are examples of possible high-value machine learning applications. Recent ground-breaking studies have shown that the use of machine learning techniques in pathology greatly enhances the ability to detect metastases in lymph nodes, grade breast cancer using the Ki67 score, grade prostate cancer using the Gleason score, and grade melanoma using the Tumour-Infiltrating Lymphocyte (TIL) score. Additionally, it has been shown that deep learning algorithms can predict the status of several molecular markers in lung, prostate, gastric, and colorectal cancer based on conventional HE slides.

Prognostic (survival outcomes) deep neural network models have also been demonstrated in a number of illnesses, including lung cancer, melanoma, and glioma. These models are based on digitized HE slides. In this review, we set out to discuss and compile the most recent advances in artificial intelligence's use in diagnostic pathology and digital picture processing. The current focus of AI research in pathology is on regular diagnosis support and prognostication, notably for cancer patients. According to preliminary research, using a computer can help pathologists make a diagnosis more quickly and correctly. The sensitivity for finding micro metastases increased from 83.3% (by a pathologist alone) to 91.2% in a pilot trial on the diagnosis of breast cancer including 70 patients (by a pathologist combined with a computer algorithm). The results also points to the possibility that AI used to analyses the histomorphological characteristics of cells under a microscope could help infer specific genetic traits including mutations in important genes and DNA methylation patterns.

Several studies have been conducted on prediction models for patient survival and loco regional recurrences in patients with oral squamous cell carcinomas in order to improve prognostication of oral cancer. Only a few researches have examined oral cancer digital computational images for machine learning techniques. It is clear that additional study at the level of the entire slide image is necessary, and computational pathology may advance as a result of potential future partnerships with computer scientists.