Abstract
Aim: The study aims to assess the efficiency of deep machine learning (convolutional neural network architecture) for the diagnosis of psammomatous meningioma by evaluating the digital pathology images.
Materials and Methods: A total of three hundred and twenty (320) digital images have been acquired from the anonymized hematoxylin and eosin-stained slides, which included 161 images of Psammomatous Meningioma and 159 images of normal intracranial tissue. The dataset was divided into a train set, 80% of the entire data, and 20% into the test data set. ResNet-18 architecture (state of the art deep learning computer vision algorithms) is used to diagnose these cases.
Results: A total of 161 images of psammomatous meningioma and 159 images of normal tissue are used; 80% of the collected data was used for training, where 20% for testing purposes. Using deep learning, we achieved 98.79 F1-Score and 98.4% accuracy.
Discussion: The advancement in the field of artificial intelligence has opened a lot of new channels and generated new opportunities to develop computer-aided diagnostic systems. The applications of machine learning have revealed promising results for the histopathological evaluation of neoplastic lesions. In the present study, the excellent diagnostic accuracy (98.4 %) has been achieved with the convolutional neural network architecture. There was an F1-score of 98.79 for the diagnosis of psammomatous meningioma in the present series. The studies conducted for the diagnosis of breast cancer metastasis, lung carcinoma, prostatic malignant tumors, and basal cell carcinoma also revealed excellent results. Convolutional neural network (CNN) architecture is emerging as a quite efficient deep machine learning tool for the analysis of the pathology images. This technology may be quite a valuable adjunct tool in diagnostic surgical pathology.