Artificial intelligence (AI) techniques have emerged as powerful tools with significant potential in healthcare and biomedicine. AI methods, including machine learning, deep learning, natural language processing, and computer vision, have been successfully employed for disease prediction, diagnosis, treatment planning, and drug discovery. Machine learning algorithms enable the identification of patterns and correlations in patient data, while deep learning models excel in medical signal/image analysis and genomics. Natural language processing techniques facilitate the study of unstructured clinical text, and computer vision algorithms enable automated interpretation of medical images. 

In biomedicine, segmentation is annotating pixels that form an important structure in a medical image. AI models can aid clinicians by highlighting pixels that may show signs of a certain disease or anomaly. 

According to Marianne Rakic, an MIT computer science PhD candidate, having options can help decision-making. Even just seeing that there is uncertainty in a medical image can influence someone's decisions. Rakic is the lead of a paper that introduces a novel AI tool that can capture the uncertainty in a medical image.

Introducing Tyche

Named after the Greek divinity of chance, the Tyche model provides multiple plausible segmentations highlighting slightly different areas of a medical image. The model's user can specify how many options Tyche outputs and select the most appropriate one for their purpose.

Tyche can tackle new segmentation tests without needing to be retained. Training is a data-intensive process that involves showing a model many examples and requires extensive ML experience. According to the report, Tyche could be more accessible for clinicians and biomedical researchers to use than some other methods.

The tool can be leveraged for a variety of tasks, from identifying lesions in a lung X-ray to pinpointing anomalies in a brain MRI. 

According to Adrian Dalca, an assistant professor at Harvard Medical School and MGH, and a research scientist in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), if a model completely misses a nodule that three experts say is there and two experts say is not, that is probably something you should pay attention to.

Tackling ambiguity

AI systems for medical image segmentation typically use neural networks. Loosely based on the human brain, neural networks are ML models comprising many interconnected layers of nodes or neurons that process data. 

After speaking with collaborators at the Broad Institute and MGH who use these systems, the researchers realized two major issues limit their effectiveness. The models cannot capture uncertainty and they must be retained for even a slightly different segmentation task. 

When the researchers tested an AI model with datasets of annotated medical images, they found that its predictions captured the diversity of human annotators and were better and faster than any of the baseline models. 

They plan to use a more flexible context set for future work, perhaps including text or multiple types of images. They are also looking forward to exploring ways that could improve Tyche's worst predictions and enhance the system so it can recommend the best segmentation candidates. 

MIT News

Image: Unsplash