隐空间实验室 LatentLab

← All demos · Identity Recognition · Physiological Age · Risk Assessment: manipulated-image detection

Treatment Response (pCR)

Predict whether breast cancer will respond completely to neoadjuvant chemotherapy, from MRI before and early in treatment.

How it works

INPUT Before DCE-MRI + tumour mask NIfTI Early in treatment + tumour mask Δtdays between scans Preprocessing (CPU) ① resample to 1 mm voxels ② z-score intensities ③ 150 mm cube around tumour ④ cohort amplitude spectrum ⑤ k-means radiomics → 3 subregions ⑥ keep tumour + 10 mm Siamese 3-D ResNet-18 (GPU) shared weights for both time points whole 3 regions + channel attention 512-d feature each Temporal Transformer change after minus before, given Δt OUTPUT P(pCR) ≥ 0.5 → pCR < 0.5 → non-pCR

Both scans are resampled to 1 mm, z-scored and cropped to a 150 mm cube around the tumour. The amplitude spectrum is replaced by a cohort average, so scanner-specific contrast is removed while the image phase (the structure) is kept. A k-means model on local radiomics features splits each tumour into three subregions. A Siamese 3-D ResNet-18 encodes the whole tumour and each subregion at both time points; a temporal transformer reads the change together with the number of days between the scans and outputs the probability of pathological complete response (pCR).

Input

  • Pre-treatment breast DCE-MRI volume (NIfTI, .nii or .nii.gz)
  • Its tumour segmentation mask (same size, tumour = 1)
  • Early-treatment DCE-MRI volume and its tumour mask
  • Days between the two scans

Output

  • Probability of pCR (≥ 0.5 → predicted pCR)
  • Slices of the preprocessed tumours with the three subregions

Tang W, Jin C, et al. eClinicalMedicine 2025; 85: 103298. External validation AUC 0.892–0.923 (5 centres, 1,044 patients). Typical run time: About 15–60 s, most of it CPU preprocessing.

Your data is not saved. Uploaded files are written only to a temporary folder on the lab GPU computer and deleted the moment the computation ends (or fails, or you cancel). Results are held in memory for 15 minutes so this page can show them, then discarded. Nothing is copied to our storage or logs.
Remove patient identity information before uploading. Strip names, IDs, birth dates, addresses, hospital and physician names from the files (for DICOM, use your PACS or a de-identification tool). This page checks DICOM headers in your browser before upload and warns you if such fields are still filled in. Research demo only — not for diagnosis or treatment decisions. Upload limit: 400 MB.

Run it

Before treatment

Click or drop a .nii / .nii.gz file
    Same size as the MRI; tumour = 1

      Early in treatment

      Click or drop a .nii / .nii.gz file
        Same size as the MRI; tumour = 1
          Preview the volumes in your browser (nothing is uploaded)

          Voxel axes as stored in the file; the mask is drawn in orange.

          Before

          Early in treatment