Healthcare & Bio AI
ResNet models detected breast cancer in ultrasound images with more than 99 percent accuracy in a 2025 study
In a 2025 paper, researchers applied variants of ResNet, a deep learning model that uses shortcut connections to help train very deep networks, to breast ultrasound images. The models classified the cases as benign or malignant and reported an accuracy above 99 percent in a 3-fold cross-validation test. Doctors commonly use breast ultrasound for breast cancer evaluation because it is safe and cost-effective.
What did the study do?
The authors describe a system that acquires and resizes breast ultrasound images, extracts features with a ResNet variant, and classifies them using a SoftMax classifier. The SoftMax step turns the model's numerical scores into probabilities for the benign and malignant decisions.
How was the model checked?
The study used 3-fold cross-validation. In this procedure, researchers split the data into three parts and test the model on each part in turn so every part serves as a test set once.
What does the accuracy number mean?
In this study, accuracy above 99 percent means the model made the correct benign or malignant classification in more than 99 percent of the cases the researchers tested. The abstract reports this result for the ResNet variants with the SoftMax classifier.
Limits and what's next
The abstract does not state limitations or next steps.
Source: K. Vijayakumar et al., “Detection of Benign/Malignent Breast Cancer with ResNet Models,” 2025.