Prostate Cancer has become one of the deadliest cancers among males in many nations. Pathologists use various approaches for the detection and the staging of prostate cancer. Microscopic inspection of biopsy tissues is the most accurate approach among them. The Gleason grading system is used to evaluate the stage of Prostate Cancer using prostate biopsy samples. The task of assigning a grade to each region in a tissue is a time-consuming task. Furthermore, this task often has several challenges since it has considerable inter-observer variability even among expert pathologists. In this paper, we propose an automatic method for this task using a deep learningbased approach. For this purpose, we use the DeepLabV3+ model with MobileNetV2 backbone and train it with the newly released dataset from Gleason 2019 challenge. Our model achieves the mean Cohen's quadratic kappa score of 0.56 with the pathologists' annotations on the test subset which is higher than the inter-pathologists' score (0.55).
Paper
Full text
Towards Automatic Prostate Gleason Grading Via Deep Convolutional Neural Networks
Semantic Scholar · Medicine · 2019
Abstract
Prostate Cancer has become one of the deadliest cancers among males in many nations. Pathologists use various approaches for the detection and the staging of prostate cancer. Microscopic inspection of biopsy tissues is the most accurate approach among them. The Gleason grading system is used to evaluate the stage of Prostate Cancer using prostate biopsy samples. The task of assigning a grade to each region in a tissue is a time-consuming task. Furthermore, this task often has several challenges since it has considerable inter-observer variability even among expert pathologists. In this paper, we propose an automatic method for this task using a deep learningbased approach. For this purpose, we use the DeepLabV3+ model with MobileNetV2 backbone and train it with the newly released dataset from Gleason 2019 challenge. Our model achieves the mean Cohen's quadratic kappa score of 0.56 with the pathologists' annotations on the test subset which is higher than the inter-pathologists' score (0.55).