Article Text

Download PDFPDF
1163 Comparison and implementation of machine learning algorithms for breast cancer recurrence prediction
  1. Ines Houissa1,
  2. Montassar Ghalleb1,
  3. Halima Jilani2,
  4. Sana Hamdi3,
  5. Imen Sassi1,
  6. Med Amine Bouida1,
  7. Yosr Zenzri4,
  8. Alia Mousli5,
  9. Yoldoz Houcine6,
  10. Ines Zemni1 and
  11. Tarak Ben Dhieb1
  1. 1Surgical Oncology Department Salah Azaiez Institute, Tunis, Tunisia
  2. 2IT Department, Central University, Tunis, Tunisia
  3. 3Faculty of Science, Université Tunis Manar, Tunis, Tunisia
  4. 4Oncoloy Department, Salah Azaiez Institute, Tunis, Tunisia
  5. 5Radiotherapy Department, Salah Azaiez Institute, Tunis, Tunisia
  6. 6Pathology Oncology Department, Salah Azaiez Institute, Tunis, Tunisia


Introduction/Background About 15% of women treated for an early-stage breast cancer(ESBC)developed recurrence within 10 years. Contemporary approaches to artificial intelligence (AI) based on deep learning have generated interest in its application to breast cancer recurrence detection.

Methodology We reviewed the clinical record of 332 patients treated in Salah Azaiez Institute(2006–2018).

Results The first step consisted in an accurate data analysis of the prognostic factors impacting the risk of disease recurrence including age, disease stage, histological subtype and grade, hormonal receptors ‘status, tumor invasion and adjuvant treatment.

As part of the data pre-processing, various replacement operations were carried out to ensure the completeness of our dataset using the SMOTE approach.

We assessed dataset size, degree of class balance, validation strategies, sample techniques,and data handling strategy, all of which have a direct influence on training and testing performance.

Once we had completed the data pre-processing steps, simulation phase was initiated to create a highly accurate model capable of reliably predicting node-negative breast cancer relapse based on its characteristics.

A variety of algorithmic techniques were applied, comparing them to determine the best-performing model.

Using the SMOTE approach, we generated new positive instances in order to balance the proportion of classes in the dataset. This data augmentation approach improved the performance of the various models.

In fact, our method had brought an additional variability that reinforces the representative and exhaustive nature of the training set.

Through this project, we have achieved promising results, paving the way for potential practical applications of AI in the healthcare field offering the opportunity for personalized care ans improved therapy response rates.

Conclusion This study provided an overview of the basic concepts and developments in the field of AI for recurrence detection of ESBC with negative axillary lymph nodes offering the opportunity for clinicians to establish a personalized care and avoid ineffective overtreatment.

Disclosures The authors have no disclosure to declare.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.