IFLYTEK neuroimaging disease prediction program!

Background of the contest

cerebral MRI(Magnetic Resonance Imaging ) The full name is brain magnetic resonance imaging , It is an imaging that reflects the pathological tissue structure of the brain , It is based on the atomic nucleus with magnetic distance under the action of magnetic field , The principle that can produce the transition between energy levels, so as to provide the biological brain structure information of disease for clinic .

Modern medical clinical practice shows that MRI It can greatly improve the accuracy of diagnosis , Thus for brain epilepsy 、 Brain tumors 、 Parkinson's disease 、 Alzheimer's syndrome and other brain diseases that are not obvious at the early stage of the disease provide an effective means of early detection . So as to delay the onset of the disease , It has positive significance for the follow-up rehabilitation treatment of patients .

Address of the competition ：http://challenge.xfyun.cn/topic/info?type=NADP&ch=ds22-dw-gzh01

Code address ：https://github.com/datawhalechina/competition-baseline

The mission of the event

For research based on the brain MRI Disease prediction for , This competition provides brain MRI Data set training samples , The brains of elderly volunteers were recorded MRI Information , These include mild cognitive impairment （MCI） The patient's brain image data 、 Alzheimer's syndrome （AD） Brain image data of patients and healthy people （NC） Brain imaging data .

The subjects were divided into three categories according to the medical diagnosis ：

• NC： health

• MCI： Mild cognitive impairment

• AD： Alzheimer's syndrome

Participants need to build models based on the samples provided , Analysis and prediction of Alzheimer's syndrome .

Data is introduced

The competition is divided into two stages: preliminary and semi-final , The difference between the two stages is the magnitude of the samples provided , And the classification tasks are different ：

• The preliminaries are AD And NC Two classification

• Second round MCI And NC Two classification

This model is based on the submitted result document , use F1-score Evaluate .

Game Modeling

Game title is a very classic image classification model , Use the existing... In the model under construction MRI Data modeling is enough . The following points should be paid attention to in this paper ：

• data fetch , Read required NII Format

• Channel selection , The original channel is greater than 3

• Model and data amplification

details 1： Read NII Format

import nibabel as nib
from nibabel.viewers import OrthoSlicer3D
img = nib.load(path)

details 2： Random channel selection

idx = np.random.choice(range(img.shape[0]), 130)
idx.sort()
img = img[idx, :, :]
img = img.astype(np.float32)

details 3： Model network structure

model = models.resnet18(True)
model.conv1 = torch.nn.Conv2d(130, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
model.avgpool = nn.AdaptiveAvgPool2d(1)
model.fc = nn.Linear(512, 2)

details 4： Data amplification method

A.Compose([
      A.RandomRotate90(),
      A.RandomCrop(128, 128),
      A.HorizontalFlip(p=0.5),
      A.RandomContrast(p=0.5),
      A.RandomBrightnessContrast(p=0.5),
])

If the crowd is full , Focus on Datawhale official account , reply “ data mining ” or “CV” or “NLP” You can be invited to join your own team group , In addition to the exchange of experience , The competition questions will also be updated and released in the group .

One key, three links , Learning together ️