Gamma knife knowledge-based planning with isocenter selection

Area of Science:

• Radiosurgery
• Medical Physics
• Deep Learning in Medicine

Background:

• Gamma Knife (GK) radiosurgery planning balances dose distribution and treatment time.
• Current methods include manual forward planning, inverse planning, and knowledge-based planning (KBP).
• Existing KBP for GK uses 3D dose prediction but relies on manual isocenter selection, limiting plan quality.

Purpose of the Study:

• To develop a comprehensive KBP pipeline for GK.
• Integrate 3D dose prediction with simultaneous optimization of isocenter locations and beam-on times.

Main Methods:

• A deep learning model generated 3D dose predictions for 20 patients.
• A mixed-integer model (GK-KBP-OptIso) optimized isocenter selection and beam-on time.
• Plans were compared against KBP with fixed isocenters (GK-KBP-FixIso) and historical clinical plans.

Main Results:

• GK-KBP-OptIso achieved comparable conformity to GK-KBP-FixIso with improved dose falloff.
• When allowed variable isocenters, GK-KBP-OptIso produced higher conformity plans than clinical plans.
• Treatment times were comparable or slightly increased compared to manual methods.

Conclusions:

• A KBP pipeline integrating isocenter selection produces GK plans equal or superior to manual KBP and clinical planning.
• This automated approach can streamline the generation of high-quality GK treatment plans.
• The method shows potential for improving efficiency and quality in radiosurgery.