Pathology's Changing Environment: Incorporating AI and Its Benefits

Until recently, the practice of pathology has been entirely “human-driven”.  Well-trained pathologists examine all tissues and arrive at diagnoses based on their application of learned criteria and experience.  However, it is well-known that the accuracy of human interpretation can be hampered by subjectivity (inter-observer variability), inconsistency (intra-observer variability), and fatigue. Recently, the rise of digital methods in pathology has led to a growing interest in applying artificial intelligence (AI) to aid or even improve on the analysis of medical specimens.  For the pathologist, AI has the potential to improve accuracy, productivity, and workflow by allowing the computer do what it does well: consume lots of data, recognize patterns, and perform automated analyses.  Objective and reproducible specimen examination, along with certainty that all material has been “seen,” leads to greater accuracy.  AI can be trained to identify specific features and present them selectively to human observers, leading to prescreening (“guided screening”) of specimens and promising improved workflow and productivity.  And, finally, the “holy grail”: under the right circumstances, AI-driven systems can make novel observations of morphologic patterns in pathology specimens, potentially leading to new knowledge and ultimately computer-generated diagnosis.  In this blog, we describe our general approach to applying AI to digital pathology specimens, detail some of the important steps involved in establishing an AI workflow, and discuss the potential benefits of AI.

AI Approach

Let us explain first what we mean by “AI,” since the term is used liberally and sometimes synonymously with related concepts such as computer vision, machine learning, and deep learning.  We define an AI algorithm as one in which the computer exhibits its own “intelligence” by “learning” the correct answer to a problem.  In medical imaging, the problem is usually to identify (detect) and interpret (classify) different phenomena in the image data.  Classifiers that use training data (data examples annotated as belonging to a set of classification categories) to identify mathematical discriminators that help determine what data belongs in what category all display a basic kind of AI.  Often, simple classifiers rely on predefined features computed from the raw data as metadata to perform the classification, in which case artificial intelligence (feature-based classification) is helped by human intelligence (definition of the features).  However, convolutional neural networks (CNNs) extend the AI concept by “learning” feature filters within their hidden layers that best classify the data.  It is the presence of multiple hidden layers (layers between the input layer [training data] and output layer [classifier output]) that define a “deep learning” neural network such as a CNN.  Our use of AI, then, indicates the application of a CNN to maximize classification performance for a digital pathology imaging problem of interest.

How to Set up an AI Workflow

There are several important steps involved in setting up an AI workflow. The first step is to define the input data into the system.  For example, one might specify that the input images to be classified are rectangular sub-images of a certain size derived from an overlapping grid applied to a whole slide image (WSI). Second, input data should be divided into training data (used to train the classifier) and test data (used to evaluate the classifier.) Third, both training data and test data should be annotated by an expert, to establish a ground truth classification category for each input image.  Fourth, the CNN architecture needs to be defined, most importantly the number and size of the convolutional layers that apply convolutional filters to derive the CNN features and the number and size of fully-connected layers that compute the weights that determine the final classification.  Additional layers may be added in constructing the CNN, including pooling layers to combine information from previous layers, “activation” layers that introduce nonlinearity into the network to improve the convergence of training, and dropout layers that reduce overfitting. Fifth, the CNN should then be trained on the training data.  This is the most time-consuming part of the process and results in a classifier that uses the annotated training data to determine what types of input images result in each classification category.  Sixth, the trained CNN should be applied to the test data, to see how the CNN classifies each example in the test data set.  And finally, the classification of the CNN on the test data should be evaluated by comparing against the ground truth annotations of the test data.  For detection problems with only two classification categories (“present” or “absent”), evaluation is in terms of sensitivity (probability that the CNN declares “present” when the ground truth annotation is “present”) and specificity (probability that the CNN declares “absent” when the ground truth annotation is “absent”).  When there are multiple classification categories, the evaluation takes the form of a results matrix (termed a confusion matrix) in which each row displays the probability of correct classification for a given category as well the probability of incorrect classification of that category in terms of all the other categories.

Potential Benefits

By producing reliable classification of medical imaging data, AI has several potential benefits. When used for detection, AI can reduce the workload for pathologists by prescreening a potentially large amount of slide data, enabling pathologists to focus on only those regions where phenomena of interest are present.  Reducing the pathologists’ workload in this way should improve productivity and reduce fatigue, leading to overall better diagnostic accuracy.  Used for classification in multi-class classification problems, AI can aid pathologists’ interpretations by indicating the type of phenomenon present (in addition to the present/absent determination in detection).  For example, one might conceive of several types of abnormalities for a given tissue structure.  Given enough training data in each class, a CNN should be able to classify the type of abnormality successfully, thereby aiding pathologists’ interpretations.  Further, by extracting the feature filters learned by the CNN, one can also visualize the features/structures the CNN uses to classify the data.  This may give insight to pathologists on how to improve their human interpretations for the future.

To find out more about digital pathology, deep learning, and AI, contact us at info@corista.com or visit our website.