{"cells":[{"metadata":{},"cell_type":"markdown","source":"","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"## Table of Contents\n1. [Introduction](#introduction)\n2. [Medical Professional input](#presentation)\n3. [Task Overview](#task_overview)\n4. [Team Overview](#team_overview)\n5. [Next Tasks / Datasets (2 weeks)](#task_next)\n6. [Daily calls](#task_calls)\n\n","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"\n\n\n# 1. Introduction \nThis is a notebook created by a collaborative effort of CoronaWhy.org, multi-disciplinary global effort of volunteers. \n\n- Visit our [website](https://www.coronawhy.org) to learn more.\n- Read our [story](https://medium.com/@arturkiulian/im-an-ai-researcher-and-here-s-how-i-fight-corona-1e0aa8f3e714).\n- Visit our [project page](https://www.coronawhy.org/projects/pulmonary-fibrosis-model) \n","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"# 2. Medical Professional Presentation \n\n\n#### Agenda:\n- Introduction from Sukhwinder Kaur and her current research/work\n- Overview of the pulmonary fibrosis challenge and associated medical imagery examples\n- Mapping out problems to specific tasks for #team-pulmonary-fibrosis-model\n- Q&A from community\n\n#### About Sukhwinder Kaur:\n- Assistant Professor, Biochemistry and Molecular Biology at University of Nebraska Medical Center\n\nhttps://www.unmc.edu/news.cfm?match=23347\n\nhttps://www.unmc.edu/biochemistry/faculty/kaur.html\n\n\n#### Aug 18, 2020 - Call Summary \n\n09:12 - Brief introduction of the presentation\n\n11:59 - Pulmonary Fibrosis defined as scarring of the lung tissue\n\n16:52 - Comparison of healthy lungs versus fibrotic lungs and the consequences\n\n18:30 - Symptoms and causes of the fibrosis (occupation, genetic, drugs, medications)\n\n20:05 - Tools used to diagnose and monitor the disease\n\n22:11 - Risk factors: older age, male gender, smoking, family history, etc.\n\n22:54 - Spirometry as the most common lung function test\n\n24:37- Aim of the Kaggle competition and present competition: Need to predict, for each week, the FVC prediction and model confidence\n\n25:07 - How to diagnose pulmonary fibrosis\n\n29:20 - Sample of CT scan images and the signs and patterns of the disease\n\n35:13 - HRCT (High-Resolution Computed Tomography)-used in diagnosing pulmonary fibrosis\n\n36:20 - Discussed articles/studies related to pulmonary fibrosis and the tests done\n\n45:35 - Discussions and Questions\n\n48:10 - Discussed the standard lung CT window used on the specific paper discussed\n\n50:58 - Suggestion: Segment entire CT scans into different regions and use different algorithms to look for characteristic patterns\n\n55:37 - Comments on the difference between Obstructive lung disease (hard to exhale) and Restrictive lung disease (hard to inhale)\n\n59:36 - Worry of long term effects of COVID-19 to people with mild symptoms- might develop pulmonary fibrosis in 5-10 years\n\n1:04:18 - Issue of gender bias in the scoring system, suggested running assessment with and without gender to determine the difference\n","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"from IPython.display import HTML\n\nHTML('')","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# 3. Task Overview \n\n#### Open Source Artificial Intelligence Model for Medical Imagery Screening\n\nA recently proposed alternative COVID-19 screening alternative is AI-powered diagnosis that is based on chest radiography images such as X-rays or computed tomography (CT) scans.\n\nThis is an Open Source, Open Science project for building a semi-supervised model that can be used for any CT Lung based task, including ARDS and any other COVID-19 related comorbidities.\n\nRead more on Notion:\nhttps://www.notion.so/Team-Pulmonary-Fibrosis-Model-9bab848371c14a0f9075faf88e454252\n","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"# 4. Team \n\nWe are cross-disciplinary team of data scientists, medical professionals and volunteers. \n\nIf you are interested in helping - please join our team here:\nhttps://www.coronawhy.org/join-the-fight\n","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"from IPython.display import HTML\n\nHTML('')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"from IPython.display import HTML\n\nHTML('')\n\n","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"# 5. Next Tasks / Datasets \n\n\n\nQuestions regarding radiology:\n- I have read some of the articles, I see certain features are considered predictors for IPV: (1) GGO (ground glass opacities), (2) honeycombing, (3) reticulation and (4) traction bronchectasis.\n- I was looking for a good explanation and examples what they are and how they appear in the HRCT but I did not and I am still in the dark. Does anyone in the team shade light on this?\n\n\nWe have following tasks:\n\n- understanding the data format of datasets we will use for semi supervised model\n\n- write scripts for extracting the data from datasets. they have different structure and root package usually have nested .zip or .gz packages. we need a script for each dataset\n\n- script for uploading dicoms to CoronaWhy dataverse\n\n- write scripts for converting all dicom data into normalized images\n\n- uploading images to CoronaWhy dataverse (i think it's a good idea to store both, raw and preprocessed data)\n\n- let me know if you'd like to help with any of those and if you need any help with it. meanwhile I will work on the same list going from top\n\n\n#### Data exploration and preparations\n\n\n# Datasets for unsupervised training\n\n---\n\n### 2019 Novel Coronavirus Resource (2019nCoVR)\n\nBy China National Center for Bioinformation. **104,009 CT slices from 1,489 patients**. The best bet. But we can extend it and combine with other datasets listed below\n\n[http://ncov-ai.big.ac.cn/download?lang=en](http://ncov-ai.big.ac.cn/download?lang=en)\n\n### MosMedData: Chest CT Scans with COVID-19 Related Findings\n\n**1110 patients, 1110 scans.** Has data with various pneumonia levels: normal lung tissue, no CT-signs of viral pneumonia, several ground-glass opacifications, ground-glass opacifications and regions of consolidation, diffuse ground-glass opacifications and consolidation as well as reticular changes in lungs\n\n[http://academictorrents.com/details/f2175c4676e041ea65568bb70c2bcd15c7325fd2](http://academictorrents.com/details/f2175c4676e041ea65568bb70c2bcd15c7325fd2)\n\n### COVID-CTset\n\n[https://github.com/mr7495/COVID-CTset](https://github.com/mr7495/COVID-CTset)\n\nThis dataset contains the full original **CT scans of 377 persons**. There are 15589 and 48260 CT scan images belonging to 95 Covid-19 and 282 normal persons, respectively\n\n### SARS-COV-2 Ct-Scan Dataset\n\n[https://www.kaggle.com/plameneduardo/sarscov2-ctscan-dataset](https://www.kaggle.com/plameneduardo/sarscov2-ctscan-dataset)\n\n1252 positive COVID-19 slices and 1230 negative CT slices. These data have been collected from real patients in hospitals from Sao Paulo, Brazil. Format: PNG\n\n### Medicalsegmentation\n\n1000 slices\n\n[https://medium.com/@hbjenssen/covid-19-radiology-data-collection-and-preparation-for-artificial-intelligence-4ecece97bb5b](https://medium.com/@hbjenssen/covid-19-radiology-data-collection-and-preparation-for-artificial-intelligence-4ecece97bb5b)\n\n[http://medicalsegmentation.com/covid19/](http://medicalsegmentation.com/covid19/)\n\n\n# Datasets with annotation for supervised training\n\n---\n\n### UCSD-AI4H/COVID-CT\n\n**Around 250 sices of covid and non covid cases. Contains CT data with annotation**. Including ARDS related patterns!\n\n## ieee8023/covid-chestxray-dataset\n\n[https://github.com/ieee8023/covid-chestxray-dataset/blob/master/metadata.csv](https://github.com/ieee8023/covid-chestxray-dataset/blob/master/metadata.csv)\n\n**84 CT slices. With annotation!** We can extract CT slices from dataset.\n\n### SIRM\n\n[https://www.sirm.org/en/category/senza-categoria-en/](https://www.sirm.org/en/category/senza-categoria-en/)\n\nAround 100 CT scans available for downloading. Looks like already included in medsegmentation. We can check it later. \n\n### kaggle/osic-pulmonary-fibrosis-progression\n\n\n### Radiopedia\n\n[https://radiopaedia.org/search?lang=us&page=6&q=pneumonia&scope=cases](https://radiopaedia.org/search?lang=us&page=6&q=pneumonia&scope=cases)\n\nWe can write scrapper to get the publicly available data. There are couple of hundred of CT slices with pneumonia\n\n### Eurorad\n\n[https://www.eurorad.org/advanced-search?search=pneumonia](https://www.eurorad.org/advanced-search?search=pneumonia)\n\nWe can write scrapper to get the publicly available data.\n\n\n# Cancer-related\n\n---\n\n- **DeepLesion**\n- LUNA16\n- Data Science Bowl 2017\n\n### Small datasets\n\n- Lung CT Segmentation Challenge 2017\n\n\n# X-rays\n\n---\n\n### Big dataset of chest x-rays\n\n14 Common Thorax Disease Categories. 112,120 frontal-view X-ray images of 30,805 unique patients.\n\n[http://academictorrents.com/details/557481faacd824c83fbf57dcf7b6da9383b3235a](http://academictorrents.com/details/557481faacd824c83fbf57dcf7b6da9383b3235a)\n\n### CheXpert: Huge dataset by Stanford and MIT\n\n500,000 images!\n\n","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"# 6. Daily Calls \n\n#### #team-pulmonary-fibrosis-model - Aug 14, 2020 - kickoff call\n\nKaggle Competition: The challenge is to use machine learning techniques to make a prediction with the image, metadata, and baseline FVC as input.\n\n04:31 - Agenda: Application of computer vision for the diagnosis of the pulmonary fibrosis\n\n05:21 - Short self-introduction from team members\n\n12:46 - Short intro from Serhiy who introduced this project to the team\n\n13:47 - Goal: To build a model trained in a semi-supervised way that can be used for any Lung CT-related task.\n\n15:44 - Impact of the project: Anyone can use the model once we trained it and released to the public\n\n16:57 - Shared link of Google's project regarding supervised learning and discussed gave an overview of the model\n\n18:00 - Brief overview of how the planned model would work\n\n22:00 - Preliminary data exploration is needed to understand what data sets the team would be dealing with\n\n25:10 - Age, Smoker or non-smoker- factors that need to be considered in segmenting data images\n\n28:51 - Discussed how to distribute computing credits and the process\n","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"from IPython.display import HTML\n\nHTML('')","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"**#team-pulmonary-fibrosis-model - Aug 25, 2020 - meeting with Keerti Bhogaraju**\n\nvideo recording: https://www.youtube.com/watch?v=yzzv1AlcfCw\n\n(you can watch at 2x speed)","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"from IPython.display import HTML\n\nHTML('')","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat":4,"nbformat_minor":4}