Assesment of a modern convNet model in the detection of breast cancer in the mexican population

This thesis presents an evaluation of a modern Convolutional Neural Network (ConvNet) model for detecting breast cancer in mammograms from the Mexican population. The study focused on implementing and testing a state-of-the-art ConvNet model, known as ConvNeXt, to assess its performance and reliability in diagnosing breast cancer. By employing the Tec-Salud dataset, which includes mammograms annotated by expert radiologists, and comparing it against the RSNA dataset, the research aimed to verify the model’s efficacy across different demographic and technological settings. The methodology involved preprocessing the images to standardize the data, followed by extensive training and validation of the ConvNeXt model. Performance metrics such as accuracy, sensitivity, specificity, and the area under the ROC curve were calculated to gauge the model's diagnostic power. Additionally, the study explored the impact of data augmentation and image normalization on model performance, emphasizing the challenges of applying AI in medical diagnostics across diverse populations. The findings revealed that while the ConvNeXt model demonstrated high accuracy and reliability, challenges such as overfitting and data bias persisted, highlighting the importance of continuous model training and validation. The study contributes significantly to the ongoing efforts in integrating AI into breast cancer diagnostics, offering insights into the potential of modern deep learning models to enhance early detection and treatment strategies in a demographically diverse patient population.

https://orcid.org/0000-0003-1361-5162