Truace
Subscribe

Intratumoral and Peritumoral Ultrasoundomics Models for Preoperative Prediction of Pathological Complete Response to Neoadjuvant Chemotherapy in Breast Cancer

Rationale and objectives This study investigated the clinical utility of a machine learning (ML) model based on ultrasoundomics features (UFs) from primary foci to predict pathological complete response (pCR) in patients with breast cancer (BC) undergoing neoadjuvant chemotherapy (NAC). Materials and methods Ultrasound images from 333 patients with BC treated with NAC were retrospectively analyzed. Patients in the internal training cohort and external validation cohort were from two different hospitals, respecti…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1853· cite & verify
Intratumoral and Peritumoral Ultrasoundomics Models for Preoperative Prediction of Pathological Complete Response to Neoadjuvant Chemotherapy in Breast Cancer
Tumors of the liver and intrahepatic bile ducts by Craig, John Redfernd, 1943- Peters, Robert L., 1927- Edmondson, Hugh A., 1906- Armed Forces Institute of Pathology (U.S.). Public domain

In brief

Researchers retrospectively analyzed ultrasound images from 333 breast cancer patients who received neoadjuvant chemotherapy, with training and external validation data drawn from two hospitals. They extracted quantitative image features from the tumor alone and from the tumor plus 3 mm and 5 mm of surrounding tissue, then compared eight machine learning algorithms to predict pathological complete response.

Predicting response before surgery could help clinicians judge whether chemotherapy is working. The best model, which included the 5 mm peritumoral zone, performed noticeably better than the tumor-only model on outside data, but performance fell from training to external testing, so how well it would work in routine care is still unproven.

Main points

  1. Of 333 patients, 119 (35.7%) achieved pathological complete response and 214 (64.3%) did not.
  2. The best tumor-only model used AdaBoost, with AUC 0.811 in training and 0.632 in external validation.
  3. Adding a 3 mm peritumoral region gave AUC 0.830 in training and 0.699 in external validation.
  4. The 5 mm model reached AUC 0.871 in training and 0.779 in external validation.

The gain

Including tissue 5 mm around the tumor raised external-cohort AUC for predicting pathological complete response from 0.632 (tumor only) to 0.779.

The rundown

The team outlined the tumor on ultrasound, then expanded the outline outward by 3 mm and 5 mm to create three image regions. Features were normalized with Z-scores and narrowed through Spearman correlation and LASSO selection before modeling.

Eight algorithms were compared on tumor-only features: support vector machine, K-nearest neighbor, random forest, Naive Bayes, logistic regression, AdaBoost, multilayer perceptron, and light gradient boosting. AdaBoost performed best and was then applied to the two peritumoral-expanded regions.

The authors used SHAP to quantify feature importance and produce waterfall charts for single-patient predictions, and decision curve analysis indicated the 5 mm model offered superior clinical benefit.

Sources

  1. Peer-reviewedAcademic Radiology2026-10-05

The debate