AI Deep Learning Techniques for Multicenter Detection and Classification of Invasive Breast Carcinoma on Whole Slide Histopathological Images

One of the most common cancers in the world, breast cancer presents a big problem for pathologists who have to make prompt and precise diagnosis. Tools for artificial intelligence (AI) and deep learning provide encouraging ways to help pathologists handle this growing demand. It is still difficult to create algorithms that are quick, dependable, and widely used in many medical facilities. This paper suggests a patch-based technique for identifying and localizing aggressive cancer in breast whole-slide images by utilizing a network of convolutional neural networks (CNN). First, a set of data from an example acquisition facility was used to train the network. After that, a calibration phase using transfer learning was used to modify the model for a new target data from a different collecting center with little more training data. Metrics like accuracy, recall, as well as precision were used to assess the model's performance on a pair of files ("test reference dataset" at "check target dataset") at the patch and slide levels. The reference dataset's patch-level precision recall, and precision improved to 96.8%, 99.5%, et 98.2%, whereas the target dataset's improved to 94.1%, 80.4%, and 77.9%. Accuracy, recall, and precision at the slide level were 82.1%, 81.6%, respectively 80.5% for the targeted dataset and 100.3%, 96.7%, as well as 95.6% for the reference dataset. These improved outcomes demonstrate the algorithm's efficacy and the calibration process's performance, guaranteeing dependable implementation across various healthcare facilities. This method demonstrates how AI-powered diagnostic tools can support conventional pathology operations while preserving high standards of precision and dependability.