Source-linked AI summary
Using data mining techniques for diagnosis and prognosis of cancer disease
Shweta Kharya
TL;DR
Breast cancer diagnosis and prognosis require reliable approaches for distinguishing benign from malignant tumors and predicting outcomes after cancer removal. This paper surveys data mining research and current work across breast cancer datasets, reporting decision trees as the best predictor in the cited conclusion. It also identifies larger datasets as needed for stronger statistical confidence in some associations.
Problem
Reliable methods are needed to distinguish benign from malignant breast tumors and predict outcomes or recurrence in breast cancer patients.
Method
The paper reviews technical and review articles and surveys data mining approaches, including decision trees, association-rule classifiers, Bayesian methods, neural networks, and logistic regression.
Results
Decision tree was reported as the best predictor, achieving 93.62% accuracy on the UCI benchmark and SEER datasets.
Takeaways & Limitations
Data mining techniques are presented as approaches for supporting breast cancer diagnosis and prognosis, with decision-tree prediction proposed for automated and web-based systems.
Takeaways & Limitations
Some reported associations require evaluation on larger datasets to achieve higher statistical confidence and assess more genes and SNPs.
Abstract
from arXiv · showhide
Breast cancer is one of the leading cancers for women in developed countries including India. It is the second most common cause of cancer death in women. The high incidence of breast cancer in women has increased significantly in the last years. In this paper we have discussed various data mining approaches that have been utilized for breast cancer diagnosis and prognosis. Breast Cancer Diagnosis is distinguishing of benign from malignant breast lumps and Breast Cancer Prognosis predicts when Breast Cancer is to recur in patients that have had their cancers excised. This study paper summarizes various review and technical articles on breast cancer diagnosis and prognosis also we focus on current research being carried out using the data mining techniques to enhance the breast cancer diagnosis and prognosis.
1. INTRODUCTION
The paper frames early, accurate breast cancer diagnosis and long-term prognosis as important medical prediction problems. It surveys data mining research that extracts patterns from medical data to support these tasks.
- Early detection is presented as the most effective way to reduce breast cancer deaths.
- Diagnosis distinguishes benign from malignant breast tumors without requiring surgical biopsy.
- Prognosis concerns long-term care and recurrence prediction after a patient’s cancer has been surgically removed.
- Data mining identifies patterns and relationships among many variables in historical medical datasets to predict disease outcomes.
- The study summarizes review and technical articles and surveys current data mining research on breast cancer diagnosis and prognosis.
2. AN OVERVIEW OF BREAST CANCER
Breast cancer is described as the most common cancer among women and a major cause of cancer death. Early diagnosis is associated with high five-year survival.
- Breast cancer affects approximately 10% of women at some stage of their lives.
- Breast cancer is the second leading cause of cancer death in women, although survival is high.
- With early diagnosis, 97% of women survive for five years or more.
- Risk factors include age, genetic risk, and family history.
3. METHODOLOGY
The paper’s methodology is a literature survey spanning medicine, computer science, and engineering. It emphasizes recent publications in these fields.
- The study surveys journals and publications from medicine, computer science, and engineering.
- The research focuses on more recent publications.
4. RESEARCH FINDINGS
The reviewed studies apply decision trees, association-rule mining, neural networks, and related data-mining methods to breast-cancer diagnosis and prognosis. Reported findings include promising classification and survivability prediction, with C5 decision trees outperforming the compared models in one large evaluation.
- Decision-tree association analysis: Decision-tree analysis identified statistically significant associations in breast-cancer data, but larger datasets were needed for stronger statistical confidence and broader gene–SNP correlations.The analysis used 32 SNPs from BRCA1, BRCA2, and TP53 together with tobacco and alcohol attributes.
- Digital mammography classification: Digital mammography experiments combined neural networks and association-rule mining for anomaly detection and classification, with both approaches reaching over 70% classification accuracy.The experiments used MIAS mammograms categorized as normal, benign, or malignant.
- Digital mammography classification: The mammography pipeline acquired and enhanced images, extracted features, and then classified them using association rules or neural networks.The Apriori algorithm discovered feature–category rules, while neural networks were trained with backpropagation using momentum and variable learning rates.
- Neural-network classification: Neural-network studies used feed-forward and backpropagation models, including a parallel implementation in which neurons were treated as independent parallel processes.The reported experiments described the parallel approach as producing efficient results and the classification results as satisfactory.
- Survivability prediction: A separate survivability analysis found preliminary data-mining results promising and reported that C4.5 performed better than Naïve Bayes and back-propagated neural networks.The comparison used three algorithms and a newer SEER database containing 482,052 records.
- Comparative survivability prediction: The survivability study compared artificial neural networks, decision trees, and logistic regression using 433,272 SEER records and 10-fold cross-validation.The dataset covered 1973–2000 and contained 72 variables.
- Logistic regression: Logistic regression predicts event odds rather than point estimates and assumes that log odds are linear in the predictor variables.In a two-class problem, odds above 50% assign a case to class 1; otherwise it is assigned to class 0.
- Comparative survivability prediction: C5 decision trees achieved 0.9362 classification accuracy, 0.9602 sensitivity, and 0.9066 specificity, outperforming the ANN and logistic-regression models evaluated.The ANN achieved 0.9121 accuracy, 0.9437 sensitivity, and 0.8748 specificity; logistic regression achieved 0.8920, 0.9017, and 0.8786, respectively.
5. CONCLUSION AND FUTURE WORK
Computer-aided diagnostic tools use data-mining classifiers for breast-cancer classification, with decision trees identified as the strongest predictor in the discussed datasets. The paper proposes extending this approach to web-based and remote-area diagnostic systems.
- Computer-aided diagnostic tools and data-mining techniques are presented as approaches for breast-cancer classification.
- Decision trees achieved 93.62% accuracy on both the UCI benchmark and SEER datasets.
- The paper identifies decision trees as the best predictor among the discussed classifiers and soft-computing approaches.
- Decision-tree prediction could support web-based automated systems that accept predictor variables for diagnostic use.
- The authors propose deploying decision-tree-based prediction in remote and rural regions to imitate human diagnostic expertise.