INTEGRATING PROSTATE SPECIFIC ANTIGEN DENSITY BIOMARKER INTO DEEP LEARNING PROSTATE MRI LESION SEGMENTATION MODELS.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate prostate cancer lesion segmentation in multi-parametric magnetic resonance imaging (mpMRI) is vital for diagnosis and biopsy guidance.
• Current deep learning models struggle with segmentation due to variations in lesion size and location.
• Improving segmentation performance is critical for better patient outcomes.

Purpose of the Study:

• To enhance prostate cancer lesion segmentation in mpMRI by integrating the prostate specific antigen density (PSAD) biomarker.
• To control lesion size prediction within deep learning models using feature-wise transformations.
• To improve the accuracy of automated segmentation for clinical applications.

Main Methods:

• Developed a deep learning model integrating the PSAD biomarker via feature-wise transformations.
• Conditioned latent space features to control lesion size prediction.
• Evaluated the model on a public dataset of 214 annotated mpMRI scans.
• Compared performance against a baseline 3D U-Net model.

Main Results:

• The model integrating PSAD demonstrated significantly improved segmentation performance.
• Key metrics such as Dice coefficient and centroid distance showed substantial enhancement.
• The integration effectively addressed challenges posed by lesion size and location variability.

Conclusions:

• Integrating the PSAD biomarker into deep learning models is an effective strategy for improving prostate cancer lesion segmentation in mpMRI.
• This approach offers a promising advancement for pre-biopsy diagnosis and targeted biopsy guidance.
• The findings highlight the potential of combining clinical biomarkers with AI for enhanced medical image analysis.