On feature extraction and classification in prostate cancer radiotherapy using tensor decompositions

Area of Science:

• Radiation Oncology
• Medical Physics
• Biostatistics

Background:

• External beam radiotherapy for prostate cancer necessitates understanding dose-volume effects on organs like the rectum and bladder.
• Minimizing radiation-induced toxicity is crucial for patient well-being and treatment adaptation.

Purpose of the Study:

• To introduce a novel method for classifying patients at risk of rectal bleeding using planned dose distributions.
• To evaluate the performance of this new method against existing classification techniques.

Main Methods:

• A Deterministic Multi-way Analysis (DMA) approach utilizing Canonical Polyadic (CP) decomposition of planned dose distributions.
• Non-rigid spatial alignment of patient anatomies followed by classification based on distance to defined subspaces.
• Validation using a leave-one-out cross-validation scheme on 99 prostate cancer patients.

Main Results:

• The proposed CP-DMA method demonstrated good specificity and sensitivity in classifying patients prone to rectal bleeding.
• CP-DMA outperformed traditional supervised and unsupervised learning methods.
• The novel approach also showed superiority compared to the Normal Tissue Complication Probability (NTCP) model.

Conclusions:

• CP-DMA offers a promising tool for predicting rectal bleeding risk in prostate cancer radiotherapy.
• This method aids in adapting treatment plans to reduce toxicity and improve patient outcomes.
• The study highlights the potential of advanced mathematical decomposition techniques in radiation oncology.