The assessment of Computer Vision Algorithms for the Diagnosis of Prostatic Adenocarcinoma in Surgical Specimens

Abstract Introduction Prostatic malignancy is a major cause of morbidity and fatality among men around the globe. More than a million new cases of prostatic cancer are diagnoses annually. The incidence of prostatic malignancy is rising and it is expected that more than two million new cases of prostatic carcinoma will be diagnosed in 2040. The application of machine learning to assist the histopathologists could be a very valuable adjunct tool for the histological diagnosis of prostatic malignant tumors. Aim & Objectives To evaluate the effectiveness of artificial intelligence for the histopathological diagnosis of prostatic carcinoma by analyzing the digitized pathology slides. Materials & Methods Eight hundred and two (802) images in total, were obtained from the anonymised slides stained with hematoxylin and eosin which included anonymised 337 images of prostatic adenocarcinoma and 465 anonymised images of nodular hyperplasia of prostate. Eighty percent (80%) of the total digital images were used for training and 20% for testing. Three ResNet architectures ResNet-18, ResNet-34, and ResNet-50 were employed for the analysis of these images. Results The evaluation of digital images by ResNet-18, ResNet-34, and ResNet-50 revealed the diagnostic accuracy of 97.1%, 98 % and 99.5 % respectively. Discussion The application of artificial intelligence is being considered as a very useful tool which may improve the patient care by improving the diagnostic accuracy and reducing the cost. In radiology, the application of deep learning to interpret radiological images has revealed excellent results. In the present study, the analysis of pathology images by convolutional neural network architecture revealed the diagnostic accuracy of 97.1%, 98 % and 99.5 % with by ResNet-18, ResNet-34, and ResNet-50 respectively. The findings of the present study are in accordance with the other published series, which were carried out to determine the accuracy of machine learning for the diagnosis of cancers of lung, breast and prostate. The application of deep learning for the histological diagnosis of malignant tumors could be quite helpful in improving the patient care. Conclusion The findings of the present study suggest that intelligent vision system possibly a worthwhile tool for the histopathological evaluation of prostatic tissue to differentiate between the benign and malignant disorders.