MDP most commonly refers to one of two things depending on the field: a Markov Decision Process in artificial intelligence and computer science, or methylene diphosphonate in nuclear medicine. Both are widely referenced by the abbreviation alone, so the meaning depends entirely on the context where you encountered it. Here’s what each one means and why it matters.
Markov Decision Process in AI
A Markov Decision Process is a mathematical framework used to model decision-making in situations where outcomes are partly random and partly under the control of a decision-maker. It’s one of the foundational concepts in reinforcement learning, the branch of AI where software agents learn to make sequences of choices to maximize some reward. Every time you read about an AI learning to play a game, navigate a robot, or optimize a recommendation system, there’s a good chance an MDP is part of the underlying math.
The core idea is straightforward: an agent exists in some state, takes an action, and transitions to a new state while receiving a reward (or penalty). The “Markov” part means the future only depends on the current state, not on how the agent got there. What happened three steps ago doesn’t matter; only where you are right now determines what happens next.
The Five Components
An MDP is formally defined by five elements:
- State space: All possible situations the agent can be in.
- Action space: All possible choices the agent can make at any given state.
- Transition model: The probability of moving from one state to another after taking a specific action. This captures the randomness in the environment.
- Reward function: The immediate payoff the agent receives for taking a particular action in a particular state.
- Discount factor: A number between 0 and 1 that controls how much the agent values future rewards compared to immediate ones. A discount factor close to 1 means the agent plans far ahead; close to 0 means it’s short-sighted.
The goal is to find a “policy,” which is a strategy mapping every possible state to the best action. A good policy maximizes the total reward the agent collects over time.
MDPs vs. POMDPs
A standard MDP assumes the agent can fully observe its current state. In many real-world problems, that’s not realistic. A self-driving car can’t see around a building, and a medical treatment AI doesn’t know everything happening inside a patient’s body. For these situations, researchers use a Partially Observable Markov Decision Process (POMDP), which adds observation probabilities and a “belief state” representing the agent’s best guess about where it actually is. POMDPs are significantly harder to solve but more realistic for complex applications like tutoring systems, healthcare decision support, and robotic navigation in uncertain environments.
Methylene Diphosphonate in Medicine
In nuclear medicine, MDP refers to methylene diphosphonate (also called medronate or medronic acid), a compound used in bone scans. It’s almost always paired with the radioactive tracer technetium-99m, forming what’s known as Tc-99m MDP. This combination is one of the most widely used radiopharmaceuticals in the world, and if your doctor ordered a bone scan, this is very likely what was injected.
MDP works because of its natural attraction to hydroxyapatite, the mineral crystal that makes up the hard structure of bone. After injection into a vein, the compound travels through the bloodstream and gets chemically absorbed into bone tissue, concentrating especially in areas where bone is actively being rebuilt or repaired. Because the technetium-99m attached to it emits gamma rays, a specialized camera can then detect exactly where the compound has accumulated, creating a map of bone activity throughout the entire skeleton.
What a Bone Scan Detects
Areas of abnormally high bone turnover light up as “hot spots” on the scan. This makes MDP bone scans useful for a wide range of conditions:
- Cancer that has spread to bone: Detecting metastatic disease is one of the most common reasons for ordering the scan, particularly for cancers of the prostate, breast, and lung.
- Fractures not visible on X-ray: Stress fractures and other subtle breaks that standard imaging misses often show up clearly on a bone scan.
- Bone infections: Osteomyelitis and other infections cause increased bone remodeling that the tracer picks up. Infection-related scans have become increasingly common, rising from about 4% of non-cancer bone scans in 1998 to over 26% in 2021.
- Arthritis evaluation: Roughly a third of non-cancer bone scans are ordered to assess joint disease.
- Other conditions: Paget’s disease, fibrous dysplasia, prosthetic joint complications, and metabolic bone diseases like those caused by kidney disease or hyperparathyroidism.
For detecting bone metastases, MDP bone scans have a sensitivity around 72% and a specificity around 81%, meaning they catch most lesions but can sometimes flag non-cancerous areas as suspicious. Newer tracers are being developed to improve on these numbers, but Tc-99m MDP remains the standard workhorse.
What the Scan Experience Looks Like
The process starts with a simple injection into a vein in your arm. Imaging then happens in phases. The first phase captures blood flow to the area of concern and takes about 60 to 90 seconds immediately after injection. The blood pool phase follows right after and extends to about 10 minutes. Then comes the main event: the delayed phase, which requires a wait of 2 to 6 hours after injection to give the tracer time to fully incorporate into bone. During this gap, you’re typically free to leave the imaging facility, and drinking extra fluids helps clear the tracer from soft tissues for clearer images.
The radiation dose from a standard Tc-99m MDP bone scan typically falls between about 2 and 5.5 millisieverts. For context, a single chest CT scan delivers roughly 7 millisieverts, and natural background radiation exposure in the U.S. averages about 3 millisieverts per year. The tracer is eliminated from the body through the kidneys over the following hours and days.