Breast cancer treatment prediction using MRI and AI

Can AI Predict Breast Cancer Treatment Success? How MRI and Machine Learning Could Revolutionize Care

Soraya Malik in Health & Wellness August 2025 • 4 min read.

"New research explores how combining multiparametric MRI with machine learning can accurately predict a patient's response to chemotherapy and survival outcomes, potentially transforming personalized breast cancer treatment."

Neoadjuvant chemotherapy (NAC) is a critical treatment for many breast cancer patients, helping to shrink tumors before surgery and improve the chances of successful breast-conserving procedures. However, NAC doesn't work for everyone, and determining who will benefit most is a major challenge in oncology.

Now, a groundbreaking study is exploring the potential of machine learning to predict how well a patient will respond to NAC. By combining multiparametric magnetic resonance imaging (mpMRI) with sophisticated AI algorithms, researchers are achieving remarkable accuracy in forecasting treatment outcomes and survival rates.

This article delves into the details of this innovative research, explaining how mpMRI and machine learning work together and what this could mean for the future of personalized breast cancer treatment. The research aims to solve - predictive modelling and personalized medicine in oncology by improving existing outcome prediction accuracy using non-invasive imaging analysis.

Decoding the Science: How MRI and Machine Learning Team Up

Breast cancer treatment prediction using MRI and AI

This study, published in Investigative Radiology, demonstrates how machine learning algorithms, when trained on data from mpMRI scans, can accurately predict a patient's likelihood of achieving a pathological complete response (pCR) – meaning no detectable cancer remains after treatment. The study goes beyond simply predicting pCR, also assessing the algorithms’ ability to forecast recurrence-free survival (RFS) and disease-specific survival (DSS).

The study involved 38 women with breast cancer who underwent mpMRI before and after two cycles of NAC. The mpMRI included:

Dynamic contrast-enhanced (DCE) MRI to assess blood flow and vessel permeability.
Diffusion-weighted imaging (DWI) to measure water movement in tissues, reflecting cellular density.
T2-weighted imaging to highlight tissue characteristics and edema.

From these images, researchers extracted 23 features per lesion, including qualitative assessments (following BI-RADS standards) and quantitative measurements (pharmacokinetic parameters from DCE-MRI and apparent diffusion coefficient [ADC] values from DWI). Then, they fed this data into eight different machine learning classifiers, including linear support vector machine, linear discriminant analysis, logistic regression, random forests, stochastic gradient descent, decision tree, adaptive boosting, and extreme gradient boosting (XGBoost), to identify which features were most predictive of treatment response and survival outcomes.

The Future of Breast Cancer Treatment: Personalized and Precise

The study’s results are promising. The machine learning models achieved high accuracy in predicting RCB class (AUC, 0.86) and DSS (AUC, 0.92) using XGBoost, and RFS (AUC, 0.83) with logistic regression. The most relevant features for predicting RCB class included changes in lesion size, complete shrinkage, mean transit time on DCE-MRI, minimum ADC on DWI, and peritumoral edema on T2-weighted imaging.

This research highlights the potential of machine learning to refine and personalize breast cancer treatment strategies. By identifying patients who are unlikely to respond to standard NAC early in the treatment process, clinicians can explore alternative therapies or adjust treatment plans accordingly. This approach minimizes unnecessary exposure to ineffective treatments and maximizes the chances of positive outcomes.

While these findings are exciting, further research is needed to validate these results in larger, more diverse patient populations. The team emphasizes the need for ongoing studies and is focused on addressing current study limitations to push the work forward.

About this Article -

This article was crafted using a human-AI hybrid and collaborative approach. AI assisted our team with initial drafting, research insights, identifying key questions, and image generation. Our human editors guided topic selection, defined the angle, structured the content, ensured factual accuracy and relevance, refined the tone, and conducted thorough editing to deliver helpful, high-quality information.See our About page for more information.

This article is based on research published under:

DOI-LINK: 10.1097/rli.0000000000000518, Alternate LINK

Title: Impact Of Machine Learning With Multiparametric Magnetic Resonance Imaging Of The Breast For Early Prediction Of Response To Neoadjuvant Chemotherapy And Survival Outcomes In Breast Cancer Patients

Subject: Radiology, Nuclear Medicine and imaging

Journal: Investigative Radiology

Publisher: Ovid Technologies (Wolters Kluwer Health)

Authors: Amirhessam Tahmassebi, Georg J. Wengert, Thomas H. Helbich, Zsuzsanna Bago-Horvath, Sousan Alaei, Rupert Bartsch, Peter Dubsky, Pascal Baltzer, Paola Clauser, Panagiotis Kapetas, Elizabeth A. Morris, Anke Meyer-Baese, Katja Pinker

Published: 2019-02-01

Everything You Need To Know

What is neoadjuvant chemotherapy, and why is it important to predict its success in breast cancer treatment?

Neoadjuvant chemotherapy, or NAC, is given before surgery to shrink tumors, which can increase the success of breast-conserving procedures. However, its effectiveness varies, presenting a significant challenge in oncology to determine which patients will benefit most. This is crucial because not all patients respond favorably to NAC, and understanding who will benefit avoids unnecessary treatment and side effects for those who won't.

How does multiparametric MRI (mpMRI) work with machine learning to predict breast cancer treatment outcomes?

The study uses multiparametric MRI, or mpMRI, which includes Dynamic Contrast-Enhanced MRI (DCE-MRI) to assess blood flow, Diffusion-Weighted Imaging (DWI) to measure water movement, and T2-weighted imaging to highlight tissue characteristics. Data from these scans are analyzed using machine learning algorithms to predict treatment outcomes, such as pathological complete response, recurrence-free survival, and disease-specific survival.

What were the results of the study using machine learning models, and which imaging features were most predictive?

The XGBoost model showed high accuracy in predicting RCB class (AUC, 0.86) and DSS (AUC, 0.92), while logistic regression was accurate in predicting RFS (AUC, 0.83). Key predictive features include changes in lesion size, complete shrinkage, mean transit time on DCE-MRI, minimum ADC on DWI, and peritumoral edema on T2-weighted imaging. These results highlight the potential of machine learning to refine predictions of treatment effectiveness.

How does this research contribute to predictive modeling and personalized medicine in oncology?

This research enhances predictive modeling and personalized medicine in oncology by using non-invasive imaging analysis to improve outcome prediction accuracy. By identifying which patients are likely to respond to neoadjuvant chemotherapy, treatment plans can be tailored to individual needs, potentially improving outcomes and reducing unnecessary treatments. This approach could minimize exposure to ineffective treatments and their associated side effects.

What specific machine learning algorithms were used, and why were multiple algorithms tested?

The study uses several machine learning classifiers, including linear support vector machine, linear discriminant analysis, logistic regression, random forests, stochastic gradient descent, decision tree, adaptive boosting, and extreme gradient boosting (XGBoost). Each of these algorithms has different strengths and weaknesses in identifying predictive features from the mpMRI data, allowing researchers to determine which model best predicts treatment response and survival outcomes. The diversity of models ensures robustness in the analysis and helps to identify the most reliable predictors.