Personalized Treatment Planning for Radiopharmaceutical Therapy

Personalized Treatment Planning for Radiopharmaceutical Therapy

1. Pre-therapy imaging (PET/CT)
2. Dose-estimation: Intensively investigated and partially solved (planar / SPECT/CT)
3. Treatment

• Dose prediction
• Dose effect
  • Treatment outcome
  • Adverse events

Vitalize the virtual patient

Physiologically-based pharmacokinetic (PBPK) model

• Simulate time-course of radioligand uptake in organs of virtual patient
• Organs & tumor: homogenous
• Simulate PET imaging using realistic PET simulator
• Dose calculation using the dose voxel kernel (DVK) method
• Simulate voxel-S-values matrices and convolved with phantoms organs

Reaction-diffusion computational simulation

Difusion + Metabolism 偏微分方程

Quantitative interpretation of the tumor microenvironment

Histology-driven convection-reaction-diffusion model

Cell Automata Model

Virtual Clone of a Patient / Whole body dynamic imaging & PBPK modelling

Spatial Transcriptomics for Precise Modelling

Accelerate Virtual Patient Clone using Artificial Intelligence

• machine learning for pre-therapy prediction of dosimetry
• development of voxel-wise pre-therapy dose prediction
  • integrate virtual patients to pretrain ai
  • PBPK-adapted deep learning

The PBPK loss term

L_{PBPK} = \sum_j [ReLU(b_{l,j}-G_{k, j}) + ReLU(G_{k, j}-b_{u,j})]\\ G_{k,j} = \frac 1 N \sum_{i=1}^NG_i I(C_i,G_i)\\ I(C_i,G_i) = \begin{cases} 1 & C_i=j\and 0 \le G_i\\ 0 & otherwise \end{cases}

Non-negative dose

$$L_{nonneg}=\frac 1 N \sum_{i=1}^NReLU(-G_i)$$

Deep learning on Dynamic Total-body PET for Pretherapy Dosimetry Prediction

• Seq2seq