how MIT and Takeda are using lasers to improve pill manufacturing

Researchers at MIT and Takeda have developed a new way to improve the production of drug tablets and powders using physics and machine learning. They learned to determine the size of particles in a mixture by illuminating them with a laser and analyzing the reflected light.

The production of tablets and powders that treat various diseases requires the isolation of the active pharmaceutical substance from the suspension and its drying. To do this, it is necessary to control the industrial dryer, mix the material and ensure that the compound acquires the desired properties for compression into a drug. This work depends largely on the observations of the operator.

Methods that make this process less subjective and more efficient are described in recent article in the journal Nature Communications, authored by scientists from MIT and Takeda. The authors of the article came up with a way to use physics and machine learning to classify the rough surfaces that are characteristic of particles in a mixture. The technique, which uses the Advanced Physics-Based Autocorrelation Estimator (PEACE), can change pharmaceutical manufacturing processes for tablets and powders, increasing efficiency and accuracy and reducing the number of failed batches of pharmaceutical products.

“Bad batches or failed steps in a pharmaceutical process are very serious,” says Allan Myerson, professor of practice in the MIT Department of Chemical Engineering and one of the study’s authors. “Anything that improves the reliability of pharmaceutical production, reduces time and increases compliance with standards, makes a big difference.”

The team’s work is part of an ongoing collaboration between Takeda and MIT launched in 2020. The MIT-Takeda program is designed to use the experience of both organizations to solve problems at the intersection of medicine, artificial intelligence and healthcare.

In pharmaceutical manufacturing, determining whether a compound is sufficiently well mixed and dried usually requires shutting down the industrial dryer and taking samples from the production line for testing. Researchers at Takeda thought that artificial intelligence could improve this task and reduce the shutdowns that slow down production. The team originally planned to use video to train a computer model that would replace a human operator. But determining which videos to use to train the model was still too subjective. Instead, the MIT-Takeda team decided to illuminate the particles with a laser during filtration and drying and measure the particle size distribution using physics and machine learning.

“We just point a laser beam at the drying surface and observe,” says Qi Han Zhang, a graduate student in the MIT Department of Electrical and Computer Science and the first author of the study.

The physical equation describes the interaction between the laser and the mixture, while machine learning characterizes the particle sizes. The process does not require the process to be stopped and started, which means the whole operation is safer and more efficient than the standard procedure, according to George Barbastatis, professor of mechanical engineering at MIT and lead author of the study.

The machine learning algorithm also does not require a large amount of data to train its task, because physics allows you to quickly train a neural network.

“We use physics to make up for the lack of training data so that we can train the neural network in an efficient way,” Zhang says. “Only a small amount of experimental data is enough to get a good result.”