Kaggle competitions on 3D volumes

Kaggle: Cervical Spine Fracture Detection

The goal is prediction fracture in whole neck 3D scans. The organizers provide 2k cases with partially annotated fractures as bounding boxes or/and pixel-wise segmentations.

Kaggle: Brain Tumor Radiogenomic Classification

The goal of this challenge is to Predict the status of a genetic biomarker important for brain cancer treatment.

With interpolation in Z dimension as it happens it is quite sparse

Each independent case has a dedicated folder identified by a five-digit number. Within each of these “case” folders, there are four sub-folders, each of them corresponding to each of the structural multi-parametric MRI (mpMRI) scans, in DICOM format. The exact mpMRI scans included are:

• FLAIR: Fluid Attenuated Inversion Recovery
• T1w: T1-weighted pre-contrast
• T1Gd: T1-weighted post-contrast
• T2: T2-weighted

The labels/targets are MGMT_value:

Experimentation

install this tooling

A simple way how to use this basic functions:

! pip install https://github.com/Borda/kaggle_vol-3D-classify/archive/refs/heads/main.zip

run notebooks in Kaggle

• 🧠 Brain Tumor Classif. ~ Lightning⚡EfficientNet3D
• 🧠 Brain Tumor Classif. ~ Lightning⚡MONAI-ResNet3D

local notebooks

• Brain tumor classification meets PT-Lightning and MONAI EfficientNet3D
• Brain tumor classification meets PT-Lightning and pre-trained ResNet3D

some results

Training progress with EfficientNet3D with training for 10 epochs > over 96% validation accuracy: