Breast cancer is the most common cancer with the highest death rate. Rapid detection and diagnosis lessens the impact of the disease. However, classify breast cancer using histopathological images -; tissues and cells examined under a microscope -; is a difficult task due to the bias in the data and the unavailability of large amounts of annotated data. Automatic detection of breast cancer using a convolutional neural network (CNN), a machine learning technique, has shown promise; but it is associated with a high risk of false positives and false negatives.
Without any confidence measures, such false predictions from CNN could lead to catastrophic results. But a new machine learning model developed by researchers at Michigan Technological University can assess the uncertainty of its predictions by classifying tumors benign and malignant, helping to reduce that risk.
In their article recently published in the journal IEEE Transactions on Medical Imaging, mechanical engineering graduate students Ponkrshnan Thiagarajan and Pushkar Kharinar and Susanta Ghosh, assistant professor of mechanical engineering and expert in machine learning, describe their new probabilistic model of machine learning, which outperforms similar models.
Any machine learning algorithm that has been developed so far will have some uncertainty in its prediction. There are few ways to quantify these uncertainties. Even though an algorithm tells us that a person has cancer, we don’t know the level of confidence in that prediction. “
From experience comes confidence
In the medical context, not knowing how reliable an algorithm is has made it difficult to trust computer-generated predictions. The current model is an extension of the Bayesian neural network -; a machine learning model that can evaluate an image and produce output. The parameters of this model are treated as random variables which facilitate the quantification of uncertainty.
The Michigan Tech model distinguishes between negative and positive classes by analyzing images, which at their most basic level are collections of pixels. In addition to this classification, the model can measure the uncertainty of its predictions.
In a medical laboratory, such a model promises to save time by classifying images faster than a laboratory technician. And, because the model can assess its own level of certainty, it can refer the images to a human expert when less confident.
But why does a mechanical engineer create algorithms for the medical community? Thiagarajan’s idea sprouted when he started using machine learning to reduce the computational time needed for mechanical engineering problems. Whether a calculation evaluates the deformation of building materials or determines whether a person has breast cancer, it is important to know the uncertainty of that calculation -; the key ideas remain the same.
“Breast cancer is one of the cancers with the highest mortality and the highest incidence,” Thiagarajan said. “We believe this is an exciting problem in which better algorithms can have a direct impact on people’s lives.”
Now that their study has been published, the researchers will extend the multiclass classification model for breast cancer. Their goal will be to detect cancer subtypes in addition to classifying benign and malignant tissue. And the model, although developed using histopathological images of breast cancer, can also be extended to other medical diagnoses.
“Despite the promise of machine learning-based classification models, their predictions suffer from uncertainties due to inherent randomness and bias in the data and the scarcity of large data sets,” said Ghosh. “Our work attempts to solve these problems and quantifies, uses and explains uncertainty.”
Ultimately, the model of Thiagarajan, Khairnar and Ghosh himself -; who can assess whether the images have high or low measurement uncertainty and identify when the images need the eyes of a medical expert -; represents the next steps in the machine learning effort.