Year: 2023
Type: conference-paper
Abstract: Mammography, considered the "gold standard" for breast mass detection, boasts approximately 90% sensitivity.However, the process of mammography reading and interpretation is time-consuming and susceptible to a significant number of false negatives and false positives.Compounding these issues is a notable shortage of radiologists specializing in breast imaging, highlighting a need for innovative solutions.This paper proposes the use of an artificial intelligence (AI) system, the Segment Anything Model (SAM), to address these challenges.SAM, a versatile image segmentation model, can generate segmentation masks in response to various prompts such as points, bounding boxes, masks, or textual descriptions.It employs a transformer-based architecture and utilizes a large-scale dataset for training.Within mammography, SAM could be deployed to automatically segment suspicious areas in the breast, thereby providing invaluable support to radiologists during diagnostic decision-making.Furthermore, SAM can be fine-tuned and trained on specific medical image datasets to enhance its performance.Our results underscore SAM's efficacy in producing accurate segmentation masks, demonstrating its potential for integration into human-in-theloop systems for annotation and decision support in medical image analysis.Nevertheless, further research and training are essential before implementing SAM in real-world medical scenarios.
Authors: Nikola Jovišić, Velibor Ilić, Slobodan Ilić, Ivana Šešić, Radiology Specialist Doctor's Office DOUTORA, Belgrade, Serbia, Aleksandra Dragičević
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4388411237
