Many women with breast cancer start treatment months before surgery. They hope the tumor shrinks enough to avoid a mastectomy or to make surgery easier. But right now, doctors often wait until the end to know if the treatment worked. That waiting can feel stressful and uncertain.
A new study suggests a smarter way to predict success much earlier. By combining MRI scans, blood tests, and tumor biology, a computer model can tell weeks into treatment whether a patient is on track. This could help doctors adjust therapy sooner and give patients clearer answers.
Breast cancer affects about one in eight women in their lifetime. Neoadjuvant therapy—chemotherapy or targeted drugs given before surgery—shrinks tumors and helps surgeons remove cancer more easily. For some patients, the tumor disappears completely. This is called a pathological complete response, or pCR. It often means a better long-term outcome.
But predicting who will reach pCR is hard. Doctors rely on tumor size changes on scans, which can be slow to appear. Blood tests and tumor markers are not always reliable. Many patients finish months of treatment only to learn the tumor did not respond as hoped. That can mean more surgery, stronger drugs, or a different plan altogether.
Here’s the twist: the body’s immune system and inflammation levels may hold clues earlier than imaging alone. Tumors interact with immune cells and blood vessels in complex ways. Capturing those signals requires more than a single snapshot.
The new approach uses a multimodal fusion model. Think of it like a team of specialists each bringing a different piece of the puzzle. One specialist reads MRI scans over time. Another tracks blood markers of inflammation. A third looks at baseline immune cells in the tumor. Together, they form a clearer picture than any one alone.
The MRI part uses deep learning, a type of computer program that learns patterns from images. It analyzes dynamic contrast-enhanced MRI scans taken early during treatment. These scans show how blood flows into the tumor. The model looks for subtle changes after just two treatment cycles—much sooner than traditional assessment.
The blood part tracks peripheral blood inflammatory indices. These are simple blood markers that reflect the body’s inflammation level. As treatment starts, inflammation can rise or fall in ways that hint at how the tumor is responding.
The tumor part uses baseline levels of tumor-infiltrating lymphocytes, or TILs. These are immune cells that have entered the tumor. Higher TILs often suggest a more active immune response against cancer. This baseline measure adds biological context to the imaging and blood data.
The study included 262 breast cancer patients who received neoadjuvant therapy. Researchers split them into a training group of 183 and a validation group of 79. They built several models: one using only early MRI features, one using only immune and inflammation data, and one combining all three.
The combined model performed best. In the validation group, it correctly predicted treatment success with an area under the curve of 0.90. That means it was highly accurate at distinguishing who would achieve a complete response. Specificity reached 95%, meaning it rarely missed non-responders.
The MRI-only model using early-treatment scans also performed well, with an area under the curve of 0.85. That was better than a model using only pre-treatment scans, which scored 0.75. This shows that changes seen after just two cycles carry strong predictive power.
The immune-inflammation model alone had an area under the curve of 0.73. It was less accurate than the combined model but still meaningful on its own. This suggests inflammation markers and baseline TILs add independent value beyond imaging.
This doesn't mean the model is ready for every clinic today.
Experts note that combining multiple data sources is a growing trend in cancer care. Imaging alone can miss biological signals. Blood tests alone can be noisy. Tumor biology alone may not reflect real-time changes. Bringing them together may offer a more complete view.
For patients, this could mean earlier conversations about treatment direction. If the model predicts a low chance of response, doctors might consider switching drugs sooner. If it predicts high chance, patients may feel more confident continuing the current plan. It could also help surgeons plan the extent of operation earlier.
But there are limits. The study was retrospective, meaning it looked back at past data rather than testing the model in real time. The validation group was smaller than the training group. And the model was tested at one center, which may not reflect broader patient diversity.
The next step is prospective testing. Researchers will need to apply the model in real clinical settings, track decisions, and measure outcomes. They will also need to check whether using the model actually improves patient care, not just prediction accuracy.
Larger trials across multiple hospitals will help confirm the results. Regulatory approval would require evidence that the model is safe, effective, and practical. Integration into existing MRI and lab workflows will also take time.
For now, the study offers a promising glimpse of how early data can guide breast cancer treatment. It shows that combining imaging, blood tests, and tumor biology may help doctors and patients make smarter choices sooner. As research continues, this approach could become part of routine care, giving more women a clearer path through treatment.