Optimizing Computer-Aided Diagnosis with Cost-Aware Deep Learning Models.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Machine learning for diagnostics

Background:

• Traditional Computer-Aided Diagnosis (CAD) models treat all misclassification errors equally.
• This approach overlooks differential costs of false negatives and false positives, leading to suboptimal medical decisions.
• Improving prediction sensitivity without compromising accuracy is crucial in medical diagnostics.

Purpose of the Study:

• To develop a novel deep learning-based CAD system incorporating a cost-sensitive parameter.
• To enhance diagnostic sensitivity in medical imaging while maintaining overall accuracy.
• To optimize CAD system performance for better patient outcomes and reduced healthcare costs.

Main Methods:

• A novel deep learning architecture was designed for CAD.
• A cost-sensitive parameter was integrated into the model's activation function.
• The methodology was validated on two distinct medical imaging datasets: LIDC and BreakHis.

Main Results:

• The proposed cost-sensitive CAD system demonstrated statistically significant increases in sensitivity.
• Sensitivity improved by 3.84% for the Lung Image Database Consortium (LIDC) dataset.
• Sensitivity improved by 5.4% for the Breast Cancer Histological Database (BreakHis) dataset, with maintained overall accuracy.

Conclusions:

• Integrating cost-sensitive parameters into deep learning CAD systems is vital for optimizing performance.
• This approach leads to improved diagnostic accuracy and sensitivity in medical imaging.
• The findings suggest a pathway to reduced healthcare costs and enhanced patient outcomes.