Machine Learning Algorithm for Distinguishing Ductal Carcinoma In Situ from Invasive Breast Cancer

Abstract

Purpose: Given that early identification of breast cancer type allows for less-invasive therapies, we aimed to develop a machine learning model to discriminate between ductal carcinoma in situ (DCIS) and minimally invasive breast cancer (MIBC). Methods: In this retrospective study, the health records of 420 women who underwent biopsies between 2010 and 2020 to confirm breast cancer were collected. A trained XGBoost algorithm was used to classify cancers as either DCIS or MIBC using clinical characteristics, mammographic findings, ultrasonographic findings, and histopathological features. Its performance was measured against other methods using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, precision, and F1 score. Results: The model was trained using 357 women and tested using 63 women with an overall 420 patients (mean [standard deviation] age, 57.1 [12.0] years). The model performed well when feature importance was determined, reaching an accuracy of 0.84 (95% confidence interval [CI], 0.76–0.91), an AUC of 0.93 (95% CI, 0.87–0.95), a specificity of 0.75 (95% CI, 0.67–0.83), and a sensitivity of 0.91 (95% CI, 0.76–0.94). Conclusion: The XGBoost model, combining clinical, mammographic, ultrasonographic, and histopathologic findings, can be used to discriminate DCIS from MIBC with an accuracy equivalent to that of experienced radiologists, thereby giving patients the widest range of therapeutic options.

Machine Learning Algorithms for Breast Cancer Diagnosis: Challenges, Prospects and Future Research Directions

“…On their part, Vy et al [38] have utilized machine learning classification model to differentiate carcinoma in Situ and minimally invasive breast cancer. However, Karatza et al [39] have proposed RF, NN [40] and Ensembles of Neural Network (ENN) for optimization of breast cancer diagnosis.…”

“…Haq et al [37] SVM Effective in high dimensional space Decision function Versatile Effective in situations where the number of measurements exceeds the number of sample Feature elimination Extraction Classification K-fold cross validation Jasti et al [12] Haq et al [37] KNN Helps to enhance classification yielded higher prediction accuracy than SVM Lazy learner (Instance-based earning) new data can be added seamlessly Does not make any assumption about data spread out Simple to implement 5-fold cross validation PCA Haq et al [37] Logistic Regression Accurate K-fold cross validation Optimized hyper-parameter Vy et al [38] Jasti et al [12] CNN Adjusts weights and biasness in breast cancer diagnostics Accuracy in image recognition problems…”

“…Relief-SVM achieved high accuracy of 99.91%, which is due to appropriate feature selection of Relief (Forest selection) algorithm and SVM predictive model. On the other hand, Vy et al [38] evaluated their model using performance metrics such as sensitivity, spec-ificity, accuracy, precision and F1 score. The developed model performed well reaching the accuracy of 0.84 (95% confidence interval 0.76-0.91), operating characteristic curve (AUC) of 0.93 (95% confidence interval 0.87-0.95), specificity of 0.75 (95% confidence interval 0.67-0.83) and sensitivity of 0.91 (95% confidence interval 0.76-0.94).…”

“…Vy et al [38] Machine learning classification Accuracy of 0.84 (95% confidence interval 0.76-0.91) Specificity of 0.75 (95% confidence interval 0.67-0.83) sensitivity of 0.91 (95% confidence interval 0.76-0.94) AUC of 0.93 (95% confidence interval 0.87-0.95) Karatza et al [39] Random [41] SVM Accuracy 97.7%…”

“…Difficult to determine optimal parameters Do not provide projections of probability straight away Jasti et al [12] Sammut et al [41] Haq et al [37] KNN Slow Difficult in predicting data High risk with larger data set High dimension need feature scaling Sensitive to noise data, missing values and outliers Sammut et al [41] Haq et al [37] Logistic Regression High computation time Vy et al [38] Jasti et al [12] CNN Slow due to frequent retraining High computational cost Need a lot of training data Karatza et al [39] Sammut et al [41] Haq et al [37] Random Forest High computation time Karatza et al [39] Neural network Time complexity Lack of hybridized classifiers Karatza et al [39] Ensembles neural network Time complexity Unable to effectively compute high dimensional task Jasti et al [12] K-means Does not guarantee high accuracy when executed in different iteration Jasti et al [12] Fuzzy C-means Expensive in terms of computational time Jasti et al [12] LS-SVM High execution time Jasti et al [12] Naïve Bayes High execution time Redundant attributes which misleads classification Jasti et al [12] Relief algorithm High execution time…”