Evaluation of Machine Learning Classification Models for False-Positive Reduction in Prostate Cancer Detection Using...

Area of Science:

• Medical Imaging
• Machine Learning in Healthcare
• Oncology

Background:

• Prostate cancer (PCa) diagnosis relies on accurate interpretation of MRI data.
• Improving the segmentation and classification of prostate lesions is crucial for effective diagnosis.

Purpose of the Study:

• To investigate and compare the performance of various machine learning (ML) and deep learning algorithms for prostate lesion segmentation and classification.
• To identify optimal ML models for enhancing existing diagnostic pipelines.

Main Methods:

• Evaluated classical ML algorithms (SVMs, RDFs, MLPs) and deep learning models (CNNs like ConvNeXt, ConvNet, ResNet).
• Utilized radiomic features with PCA or mRMR feature selection.
• Compared performance on whole images and segmented regions (gland, peripheral zone, lesions), including transfer learning approaches.

Main Results:

• Assessed the efficacy of different ML and deep learning architectures in binary classification of benign and malignant prostate tissues.
• Compared various optimization strategies for segmentation and classification tasks.
• Provided an exhaustive examination of ML model applicability in PCa diagnosis.

Conclusions:

• The study offers insights into the performance of diverse ML approaches for prostate cancer detection.
• Identified preferred ML models or families of models for optimizing upstream filtering in diagnostic pipelines.
• Aimed to guide the selection of the best-suited ML model for improving prostate MRI analysis.