iPCa-Net: A CNN-based framework for predicting incidental prostate cancer using multiparametric MRI.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Incidental prostate cancer (iPCa) is an early-stage, asymptomatic form of clinically significant prostate cancer (csPCa).
• Detecting iPCa is challenging due to subtle MRI differences and imbalanced datasets.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) framework, iPCa-Net, for predicting iPCa from prostatic MRIs.
• To address challenges of subtle image differences and sample imbalance in iPCa prediction.

Main Methods:

• Proposing iPCa-Net, a CNN framework designed for joint prostate transition zone segmentation and iPCa prediction.
• Utilizing a dataset of 9536 prostatic MRI slices from 448 patients with benign prostatic hyperplasia (BPH).
• Comparing iPCa-Net against eight segmentation and nine prediction state-of-the-art methods.

Main Results:

• iPCa-Net achieved superior performance in both tasks compared to existing methods.
• Outperformed the top segmentation method by 1.23% in mIoU.
• Surpassed the top prediction method by 2.06% in F1 score.

Conclusions:

• iPCa-Net demonstrates significant potential for early and accurate identification of iPCa patients.
• This advancement can greatly benefit disease management and patient outcomes.
• The model's joint optimization approach effectively handles segmentation and prediction challenges.