Using multi-label ensemble CNN classifiers to mitigate labelling inconsistencies in patch-level Gleason grading.

Area of Science:

• Computational pathology
• Medical image analysis
• Machine learning in oncology

Background:

• Accurate Gleason grading of prostate cancer is crucial for patient prognosis and treatment decisions.
• Patch-level Gleason grading in histopathology images is challenged by label inconsistencies within datasets like SICAPv2.
• Existing methods often struggle with the inherent variability and noise in histopathological data.

Purpose of the Study:

• To develop and validate a novel multi-label ensemble deep-learning approach for enhanced patch-level Gleason grading.
• To mitigate the impact of label inconsistencies in prostate histopathology datasets.
• To improve the accuracy and consistency of prostate cancer grading compared to state-of-the-art methods.

Main Methods:

• A multi-label ensemble deep-learning classifier was proposed, integrating three one-vs-all deep learning models.
• Transfer learning was employed to fine-tune a ResNet18 Convolutional Neural Network (CNN) classifier.
• An extensive ablation study was conducted to select the optimal CNN architecture.

Main Results:

• The multi-label ensemble classifier demonstrated superior performance over traditional single-label classifiers.
• Accuracy and F1-score improvements of at least 14% and 4%, respectively, were achieved.
• The proposed method effectively addressed label inconsistencies, leading to more reliable Gleason grading.

Conclusions:

• The developed multi-label ensemble deep-learning approach significantly enhances the accuracy of patch-level Gleason grading.
• This machine learning strategy offers a promising solution for improving prostate cancer diagnosis and prognosis.
• Addressing label inconsistencies is vital for advancing computational pathology in cancer grading.