EXPLORE PUBLICATIONS BY COUNTRIES

Breast Cancer, Convolutional Neural Network (CNN) , DenseNet121, Transfer Learning.

Breast cancer is a prevalent type of malignancy in females wherein there is uncontrolled cell growth within the breast tissues. Proper identification and classification are the basis for effective treatment and management. There has been potential in increasing classification accuracy as well as support for early diagnosis through more recent advancements with deep learning models, particularly when utilized in medical imaging. This research aims to enhance the precision of breast cancer classification by comparing deep learning model performance. Python and deep learning frameworks were employed in developing and comparing models for breast cancer classification using the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) dataset, which includes Digital Imaging and Communications in Medicine (DICOM) mammography images obtained through Kaggle. The data was quality-assured and made uniform. A conventional Convolutional Neural Network (CNN) was first applied for binary classification. Transfer learning was implemented with the DenseNet121 model that was pre-trained on ImageNet to improve performance. Layers of the model were frozen, and classification layers were included as custom. Fine-tuning was accomplished by unfreezing certain layers to enhance the ability of the model to discriminate between malignant and benign cases. The conventional CNN model achieved accuracy of 51.87%, weighted F1- score of 0.35, precision of 0.27, and recall of 0.52. Following transfer learning with DenseNet121, accuracy was improved to 71%, weighted F1-score of 0.71, Specificity of 0.83, Sensitivity of 0.61 and AUC of 0.7. Fine-tuning resulted in an end accuracy of 88%, with weighted F1-score, Sensitivity of 0.87, Specificity of 0.82, precision at 0.87 and Area Under the Curve (AUC) at 0.85.This study highlights the effectiveness of DenseNet121 combined with transfer learning for improving breast cancer classification accuracy using DICOM images from the CBIS-DDSM dataset, contributing to more reliable early detection and treatment strategies.