Automated Molecular Subtyping of Breast Carcinoma Using Deep Learning Techniques

Niyas, S.; Bygari, Ramya; Naik, Rachita; Viswanath, Bhavishya; Ugwekar, Dhananjay; Mathew, Tojo; Kavya, J; Kini, Jyoti R.; Rajan, Jeny

doi:10.1109/jtehm.2023.3241613

Cited by 9 publications

(1 citation statement)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, deep learning can accurately classify breast cancer molecular subtypes by analyzing complex information such as gene expression data. A study utilizing a deep learning model to analyze gene expression pro les of breast cancer samples successfully divided the samples into different molecular subtypes, providing important evidence for precision medicine[24][25][26]. Our study successfully reduced the dimensionality of high-dimensional features to 32 features most relevant to lymphatic metastasis through PCA and feature selection methods.…”

mentioning

confidence: 97%

Deep Learning-Enabled Ultrasound Radiomics for Accurate Prediction of Breast Cancer Lymph Node Metastasis

Wang,

Huang,

et al. 2024

Preprint

View full text Add to dashboard Cite

Background: Lymph node metastasis serves as a pivotal prognostic marker in breast cancer progression. The present study endeavors to devise a deep learning-driven ultrasound radiomics model for precise forecasting of lymph node metastasis in breast cancer patients. Methods: A retrospective analysis was conducted on clinical and ultrasound imaging data of breast cancer patients diagnosed surgically and pathologically at our institution between January 2018 and January 2023. The dataset was randomly stratified into training and testing subsets at a 7:3 ratio. Initially, tumor ultrasound images of breast cancer patients were annotated. Subsequently, a pre-trained Densenet121 convolutional neural network(CNN) was employed to extract intricate features from the annotated images. Principal component analysis (PCA) and feature selection techniques were implemented to diminish the dimensionality of the extracted features. These features were then consolidated and optimized using various machine learning algorithms to predict lymph node metastasis. The optimal algorithm was chosen to estimate the probability of lymph node metastasis for each patient utilizing the ultrasound radiomics model. Univariate and multivariate analyses were further conducted on clinical features to identify independent predictors of lymph node metastasis in breast cancer. Ultimately, these clinical predictors were integrated with the prediction probability of the ultrasound radiomics model to formulate a clinical-ultrasound fusion model, whose predictive accuracy was assessed on the testing subset. Results: The ultrasound radiomics model grounded in deep learning exhibited remarkable performance in forecasting lymph node metastasis in breast cancer. Among the tested algorithms, Logistic Regression (LR) outperformed its counterparts, attaining an AUC (95%CI) of 0.823 (0.775-0.872), along with a sensitivity of 0.835 and specificity of 0.699. Notably, the clinical-ultrasound fusion model further enhanced the predictive accuracy, achieving an accuracy of 0.796 in the training set and 0.820 in the testing set. Moreover, the AUC (95%CI) values were 0.863 (0.837-0.890) and 0.885 (0.847-0.922) for the training and testing sets, respectively, with corresponding sensitivities of 0.744 and 0.741, and specificities of 0.854 and 0.912. Conclusion: This study successfully developed and validated a deep learning-based ultrasound radiomics model, which exhibited high predictive accuracy and stability in forecasting lymph node metastasis in breast cancer. This model provides clinicians with valuable insights, enabling them to make more personalized and informed treatment decisions for breast cancer patients.

show abstract

mentioning

confidence: 97%