VGG16-MCA UNet improves tumor segmentation in MRI brain scans
VGG16-MCA UNet: Whole-Tumor Segmentation in 2D FLAIR MRI with Decoder-Side Channel Attention
Computer Vision and Pattern Recognition
Summary
Brain tumor segmentation in MRI images helps doctors diagnose and plan treatment, but tumors vary a lot and annotated data is limited. The authors combined a known image recognition model (VGG16) with a new attention mechanism in a network to better highlight tumor features. They trained and tested their model on two datasets using only one type of MRI scan and showed high accuracy in identifying tumor areas. Their evaluation method is more optimistic since training and testing used slices from the same patients, so the results represent an upper bound rather than performance on new patients.
What this means in practice
- •For medical imaging engineers: Build tumor segmentation tools for MRI scans that combine pretrained encoders with attention modules to improve detection on known patients.
- •For software teams developing neuroimaging tools: Integrate a precise, reproducible 2D FLAIR-based segmentation baseline into clinical or research MRI pipelines for brain tumor analysis.
- •For industrial mri device manufacturers: Potentially enhance tumor detection features in MRI scanners by embedding efficient attention-based segmentation models.$Commercial implications: Enables improved tumor segmentation software that can be marketed with MRI hardware for better diagnostic support.
Authors
Shubham Gajjar, Deep Joshi, Avi Poptani, Vishal Barot
Abstract
Automated brain tumor segmentation supports diagnosis, treatment planning, and monitoring of disease progression, but building models that generalize across heterogeneous tumors and limited annotated data remains difficult. We present VGG16-MCA UNet, a hybrid architecture pairing an ImageNet-pretrained VGG16 encoder with a decoder in which a Multi-Channel Attention (MCA) module recalibrates features after each skip-connection fusion, trained with the Focal Tversky loss to counter severe foreground-background imbalance. We evaluate the model as a 2D, FLAIR-only, whole-tumor segmenter on tumor-positive slices from two public datasets: the BraTS 2020 benchmark and the LGG MRI Segmentation dataset. Using 5-fold cross-validation and a single network formed by averaging the weights of the five fold models, the method attains an aggregate pixel-level Dice (F1) of 95.10% on our held-out BraTS 2020 split and 88.32% on LGG. These scores are computed over all test pixels pooled into a single confusion matrix rather than averaged per case, and are therefore not directly comparable to the per-case mean Dice used in the BraTS challenge protocol. All partitions were drawn over individual slices rather than over patients, so every patient contributes slices to both training and test; the figures above therefore measure interpolation within known patients and should be read as an upper bound rather than as generalization to new ones. The model segments a 256x256 slice in 66.32 ms on a single 6 GB NVIDIA RTX 2060, approximately 8 ms more than an equivalent VGG16-UNet without MCA. We release the split records and report the protocol in full, with the aim of providing a precisely specified and reproducible 2D FLAIR baseline.