Brain Tumor Segmentation and Volumetry

Abstract

Measuring a brain tumor still depends on experts tracing it by hand, slice by slice. This team adapted an ImageNet-pretrained hybrid CNN–Vision Transformer to four-sequence brain MRI, trained it on a patient-level split stratified by tumor volume, and evaluated it once on 73 held-out patients. Whole-tumor overlap with expert labels reached a Dice score of 0.848, matching validation, compared with 0.48 for a model trained from scratch. The model was deployed as a web tool that segments, measures and exports tumor volume. Results come from a public research dataset and are not intended for clinical use.

Keywords · brain MRI · tumor segmentation · volumetry · transfer learning · Vision Transformer · held-out evaluation

The team

Who did this work

3 students · 3 universities · 3 different majors

Student A

Biological Sciences / Bioengineering · rising sophomore

Peking University

Role in this project: Led the core pipeline: audited the data, designed the patient split, adapted the pretrained model to four-sequence MRI, ran training and the held-out evaluation, then designed and deployed the web tool.
Background: Fourth AI+X project in five months.

Student B

Data Science · rising junior

Minzu University of China

Background: Came in from machine learning on biological data: in an earlier AI+X project, built classifiers that tell breastfed from formula-fed infants using gut-microbiome data.

Student C

Music · rising junior

New England Conservatory of Music

Background: A cellist who came to AI+X with no prior programming background.

Students are not named. Year of study is as of the August 2026 project. Roles come from the team's final presentation and their own project write-ups.

Background

1Why measure tumors automatically?

To track a brain tumor, a clinician traces its border slice by slice across dozens of MRI images. It is slow, and two experts rarely draw exactly the same line.

In drug trials, a change in tumor volume is one of the clearest signs of whether a treatment works. A tool that outlines tumors consistently turns "visible on a scan" into a number that can be compared over time.

Method

2From raw scans to a working tool

Audited the data first.484 patients from the Medical Segmentation Decathlon brain tumor task [1], each with 155 image slices and four MRI sequences. Tumor sizes ranged from 7 to 318 mL, so the team made sure every group would contain small, medium and large tumors.
Split patients, not images.Each patient appears in only one group. The test group was locked away until the very end.
Built a from-scratch baseline.Starting from a transformer script the Project Lead provided, a teammate trained a model from zero. It reached only 0.48 overlap, which set a clear bar to beat.
Adapted a pretrained model.The team took a hybrid ResNet–Vision Transformer [2] pretrained on millions of everyday images and modified its input layer to read all four MRI sequences at once.
Shipped it as a web tool.The team deployed the model on Hugging Face: upload a scan, inspect the outline in 2D and 3D, read the volume, and export the result.

Figure 1. How the 484 patients were divided. Redrawn from the team's final presentation.

Team slide: patient split — **Figure 2.** The team's own summary of the frozen, volume-stratified split. From the final presentation.

Results

3Pretraining made the difference, and the result held on new patients

Figure 3. Overlap with radiologists' outlines (Dice). Validation and test scores match, so the model did not just memorize its training patients. Redrawn from the team's results.

**Table 1.** Key results reported in the team's final presentation.
Model	Evaluated on	Mean Dice [3]	Median volume error
From-scratch ViT baseline	Validation	0.48	—
Pretrained CNN–ViT	Validation (72)	0.848	8.2%
Pretrained CNN–ViT	Held-out test (73)	0.848	8.7%
Four-label variant	Held-out test, whole tumor	0.845	8.9%

Team slide: segmentation result — **Figure 4.** On a held-out patient: input scan, expert label, model prediction and overlay. From the final presentation.

A judgment call worth noting. The team also tried splitting each tumor into three sub-regions. The whole tumor still scored 0.845, but the sub-regions averaged only 0.585. They concluded the whole-tumor model is the one to trust for measuring volume.

Limitations

4What this does not show

It is a research demo, not a clinical tool.
All results come from a public research dataset (Medical Segmentation Decathlon) and have not been clinically validated.
Sub-region outlines are not yet reliable; volume should be read from the whole-tumor result.

From the team's presentation

5Selected slides

Team slide: web tool — **Figure 5.** The team's slide on the deployed web tool, with its Hugging Face address at the bottom. Open the live demo

Team slide: volume error — **Figure 6.** Predicted minus true tumor volume on held-out patients.

Skills and fit

6What this work shows the team can do

Data auditingLeak-free evaluation designTransfer learning3D medical image segmentationModel comparisonDeploying a usable tool

Project Lead

Researcher in Medical Imaging and Machine Learning

Senior Imaging Scientist at Novartis. MIT PhD in ultrasound imaging and medical robotics. Co-founded two medical startups.

The team ran the experiments and built the model and the web tool. The Project Lead met with the team weekly and provided a starting transformer script.

Good fit for applications in

Medical imaging · AI · Biomedical engineering

A working tool reviewers can understand at a glance makes a strong portfolio piece.

References

Antonelli, M. et al. (2022). The Medical Segmentation Decathlon. Nature Communications, 13, 4128.
Dosovitskiy, A. et al. (2021). An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale. ICLR 2021.
Dice, L. R. (1945). Measures of the Amount of Ecologic Association Between Species. Ecology, 26(3), 297–302.

Cite this outcome

AI+X team project (2026). Brain tumor segmentation and volumetry with CNN-ViT. AI and Computer Vision in BioTech - Novo Nordisk Project, On-Campus Experience, August 2026. blendedlearn.org/outcome-brain-tumor-segmentation

Upload a brain MRI. Get the tumor outlined and its volume measured.