Healthcare & Bio AI
A deep-learning model with fewer trainable parameters classified breast ultrasound images more accurately than a standard method, a 2025 study reports
In a 2025 paper, researchers report that a deep-learning model called INLCL classified breast ultrasound images into benign and malignant groups more accurately than a Support Vector Machine, a standard classification technique. The researchers evaluated the model on 698 benign and 437 malignant ultrasound cases that experienced medical professionals annotated. The model uses fewer trainable parameters than typical pre-trained deep learning models, which the authors say lowers the computing load.
Why did the researchers design a smaller model?
Breast cancer often remains undetected in many women until an advanced stage. Early detection is important for effective treatment and better patient outcomes. Earlier studies used pre-trained deep learning models, but those models often have many layers and parameters and need substantial computing resources.
How did the researchers test the model?
The researchers compared a new model, Intensive Neural Learning with Classification Logic (INLCL), with a Support Vector Machine. They used cross-validation. The dataset contained 698 benign and 437 malignant breast ultrasound images that experienced medical professionals annotated.
What did the results show?
The INLCL model outperformed traditional approaches, including SVM, in accuracy and other performance metrics. The abstract does not give exact numeric scores for those metrics.
What are common questions about this study?
What does the model do?
INLCL classifies ultrasound images into benign and malignant categories.
What does "Support Vector Machine" mean?
A Support Vector Machine is a standard classification technique that the researchers used as the comparison method.
What does "fewer trainable parameters" mean?
It means the model has fewer adjustable values than typical pre-trained deep learning models. That can reduce the computing resources the model needs.
Limits and what's next
The study used one dataset of 698 benign and 437 malignant ultrasound images, all annotated by experienced medical professionals. The abstract does not describe where the images came from. The researchers compared INLCL with a Support Vector Machine through cross-validation. This is one study, and the abstract reports no independent replication or external validation. The finding is tied to this dataset and this comparison.
Source: G. Ramkumar et al., “Advanced Deep Learning Approach for Breast Cancer Detection Using Ultrasound Images,” 2025.