Enhancing Breast Cancer Diagnosis: A Transparent, Reproducible Workflow Using CBIS-DDSM and Advanced Machine Learning Techniques

Improving Breast Cancer Diagnosis with AI **Key Challenges in Breast Cancer Diagnosis** To improve breast cancer diagnosis, access to mammography data and advanced machine learning techniques is crucial. However, researchers face several challenges: - Limited access to private datasets - Selective image sampling from public databases - Incomplete code availability These challenges make it hard to reproduce and validate results, slowing progress. In 2022, breast cancer led to 670,000 deaths worldwide. While technologies like tomosynthesis help with screening, they still produce false positives and inconsistent interpretations by radiologists, which can increase patient anxiety and healthcare costs. **Innovative Solutions** Researchers from Biomedical Deep Learning LLC and Washington University in St. Louis have developed a pilot codebase to simplify breast cancer diagnosis. This solution includes: - Image preprocessing - Model development - Evaluation They found that using larger input sizes improves the accuracy of detecting malignancies using the CBIS-DDSM mass subset, which provides detailed images and regions of interest (ROIs). **Data Processing and Model Training** The CBIS-DDSM dataset contains publicly available mammography images, carefully curated with updated segmentation and pathology labels. Key features of the codebase include: - Conversion of images from DICOM to PNG format - Enhanced focus on abnormal regions through image transformations - A customized convolutional neural network for model training Performance metrics like accuracy, precision, recall, F1 score, and AUROC are used for validation, ensuring optimal results for future research. **Insights from the Study** The research used the CBIS-DDSM mass subset, which includes 1,696 abnormal ROIs and 1,592 full mammograms. Each image was standardized and enhanced to highlight abnormalities. Key findings include: - Larger images significantly improve the detection of malignant cases. - ResNet-50 models outperformed Xception models in accuracy. - Larger images retain more detail, aiding in identifying specific cancer features. **Conclusion and Future Directions** Breast cancer screening has advanced, but challenges remain due to inconsistent methods. This study provides a clear and reproducible codebase to support model development and validation. By focusing on input size and quality control, researchers aim to improve model accuracy and reliability, promoting transparency and speeding up advancements in the field. **Get Involved** If you're interested in enhancing your company with AI, consider these steps: 1. Identify automation opportunities 2. Define key performance indicators (KPIs) 3. Select an AI solution 4. Implement gradually For advice on managing AI KPIs, contact us. Stay tuned for insights on leveraging AI in your operations!