{ "id": 275538, "title": "Notebook for tumor object detection", "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/275538", "author_name": "", "post_date": "2021-09-30T18:59:54.855125400Z", "votes": 14, "comment_count": 1, "views": 0, "content": "

I made yet another MR processing notebook :p~

\n

This one does object detection in all three planes. It only does T1wCE series (for now), but you'll get the idea.

\n

Notebook -> https://www.kaggle.com/davidbroberts/brain-tumor-object-detection

\n

If coupled with a good pre-processing routine (eliminate all black images, irregular peripheral images, non-brain images etc), it could be effective at sorting out images with tumors.

\n

I trained pytorch models on around 400 images from each plane that I hand labeled using https://makesense.ai

\n

The weights are imported into YOLOv5 OD .. YOLO also performs classification on the tumors, but the accuracy is average like most of the other models in this comp.

\n

Dataset -> https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od/

", "messages": [ { "id": "1529920", "postDate": "09/30/2021 18:59:54", "content": "

I made yet another MR processing notebook :p~

\n

This one does object detection in all three planes. It only does T1wCE series (for now), but you'll get the idea.

\n

Notebook -> https://www.kaggle.com/davidbroberts/brain-tumor-object-detection

\n

If coupled with a good pre-processing routine (eliminate all black images, irregular peripheral images, non-brain images etc), it could be effective at sorting out images with tumors.

\n

I trained pytorch models on around 400 images from each plane that I hand labeled using https://makesense.ai

\n

The weights are imported into YOLOv5 OD .. YOLO also performs classification on the tumors, but the accuracy is average like most of the other models in this comp.

\n

Dataset -> https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od/

", "rawMarkdown": "I made yet another MR processing notebook :p~\n\nThis one does object detection in all three planes. It only does T1wCE series (for now), but you'll get the idea.\n\nNotebook -> https://www.kaggle.com/davidbroberts/brain-tumor-object-detection\n\nIf coupled with a good pre-processing routine (eliminate all black images, irregular peripheral images, non-brain images etc), it could be effective at sorting out images with tumors.\n\nI trained pytorch models on around 400 images from each plane that I hand labeled using https://makesense.ai\n\nThe weights are imported into YOLOv5 OD .. YOLO also performs classification on the tumors, but the accuracy is average like most of the other models in this comp. \n\nDataset -> https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od/", "votes": null }, { "id": "1533459", "postDate": "10/04/2021 03:48:18", "content": "

I made a simple YOLO OD training notebook and dataset for all three planes of the T1wCE series that goes with this.

\n

The dataset contains ~400 hand labeled tumor JPEGs for each plane .. with classification labels and bounding box coordinates.

\n

Train Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od-train

\n

Train Datasets: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection-datasets
\nInfer Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection

", "rawMarkdown": "I made a simple YOLO OD training notebook and dataset for all three planes of the T1wCE series that goes with this.\n\nThe dataset contains ~400 hand labeled tumor JPEGs for each plane .. with classification labels and bounding box coordinates.\n\nTrain Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od-train\n\nTrain Datasets: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection-datasets\nInfer Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection", "votes": null } ], "comments": [ { "id": 1533459, "author_name": "davidbroberts", "author_url": "", "post_date": "10/04/2021 03:48:18", "content": "

I made a simple YOLO OD training notebook and dataset for all three planes of the T1wCE series that goes with this.

\n

The dataset contains ~400 hand labeled tumor JPEGs for each plane .. with classification labels and bounding box coordinates.

\n

Train Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od-train

\n

Train Datasets: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection-datasets
\nInfer Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1529920": "I made yet another MR processing notebook :p~\n\nThis one does object detection in all three planes. It only does T1wCE series (for now), but you'll get the idea.\n\nNotebook -> https://www.kaggle.com/davidbroberts/brain-tumor-object-detection\n\nIf coupled with a good pre-processing routine (eliminate all black images, irregular peripheral images, non-brain images etc), it could be effective at sorting out images with tumors.\n\nI trained pytorch models on around 400 images from each plane that I hand labeled using https://makesense.ai\n\nThe weights are imported into YOLOv5 OD .. YOLO also performs classification on the tumors, but the accuracy is average like most of the other models in this comp. \n\nDataset -> https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od/", "1533459": "I made a simple YOLO OD training notebook and dataset for all three planes of the T1wCE series that goes with this.\n\nThe dataset contains ~400 hand labeled tumor JPEGs for each plane .. with classification labels and bounding box coordinates.\n\nTrain Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-yolo-od-train\n\nTrain Datasets: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection-datasets\nInfer Notebook: https://www.kaggle.com/davidbroberts/brain-tumor-object-detection" }, "source": "meta" }