liveBook · Manning

Organ at risk

As explained in Applied Reinforcement Learning

An organ at risk is a healthy organ near a radiation target whose exposure must be kept below a clinical dose limit.

An organ at risk (OAR) is a healthy organ near a radiation target that may receive radiation during treatment. Depending on the tumor’s location, OARs can include the heart, lungs, spinal cord, or brain. Excessive exposure can cause serious side effects. section 8.4

Treatment planning balances delivering enough radiation to control the tumor against keeping OAR doses below safe limits. Because dose accumulates across treatment fractions, an early beam decision can constrain later choices. section 8.4

Tracking OARs in a treatment simulation

A simulated treatment environment tracks each OAR’s cumulative dose and limit. Its state includes the remaining margin, calculated as (limit − dose) / limit. The reward penalizes dose above a limit, weighted by organ risk; separate checks use configurable margins to determine success or failure. section 8.4

Listing 8.2 Constructing the state representation
def _get_state(self):
    cum_norm = self.cum_doses / 100.0  #A
 
    tumor_dose = self.cum_doses[0]  #B
    tumor_remaining = (self.target_dose - tumor_dose)
    ➥/ self.target_dose  #B
    tumor_remaining = np.clip(tumor_remaining, -1.0, 1.0)  #B
 
    oar_doses = self.cum_doses[1:]  #C
    oar_limits = self.organ_limits[1:]  #C
    oar_margins = (oar_limits - oar_doses) / oar_limits  #C
    oar_margins = np.clip(oar_margins, -1.0, 1.0)  #C
 
    tumor_size_norm = (
        (self.tumor_size - self.min_tumor_size)
        / (self.max_tumor_size - self.min_tumor_size)
    )  #D
    tumor_size_norm = np.clip(tumor_size_norm, 0.0, 1.0)  #D
 
    frac_progress = self.current_fraction
    ➥/ float(self.n_fractions)  #E
 
    state = np.concatenate(
        [
            cum_norm,
            np.array([tumor_remaining], dtype=np.float32),
            oar_margins,
            np.array([tumor_size_norm], dtype=np.float32),
            np.array([frac_progress], dtype=np.float32),
        ]
    )  #F
    return state.astype(np.float32)

The state representation calculates an OAR margin for each simulated organ, alongside tumor dose and treatment progress. section 8.4

How Applied Reinforcement Learning covers it

How deep it goes
Defines OARs and implements dose margins, weighted penalties, and success checks in a treatment simulation.
Examples
Uses the heart, lungs, spinal cord, and brain as possible OARs and a simulated multi-organ treatment environment.
What you can do after
Evaluate and understand how simulated treatment decisions balance tumor dose against healthy-organ exposure.
How it connects
Connects radiotherapy, cumulative dose, reinforcement learning, state representation, and reward signals.

This explainer was generated by Manning's knowledge graph from the text of Applied Reinforcement Learning. The section links point to the parts of the book it draws on.