<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Bartlomiej Kuchnowski on Medium]]></title>
        <description><![CDATA[Stories by Bartlomiej Kuchnowski on Medium]]></description>
        <link>https://medium.com/@bartlomiej.kuchnowski?source=rss-19daa8764b45------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*RR5jXpdNACULcToUn1X4lg.png</url>
            <title>Stories by Bartlomiej Kuchnowski on Medium</title>
            <link>https://medium.com/@bartlomiej.kuchnowski?source=rss-19daa8764b45------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 20:54:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bartlomiej.kuchnowski/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AI in ER: How Artificial Intelligence can transform Emergency Medicine as we know it.]]></title>
            <link>https://medium.com/tooploox-ai/ai-in-er-how-artificial-intelligence-can-transform-emergency-medicine-as-we-know-it-7db9c1a803d5?source=rss-19daa8764b45------2</link>
            <guid isPermaLink="false">https://medium.com/p/7db9c1a803d5</guid>
            <category><![CDATA[medicine]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[emergency-medicine]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[healthcare-technology]]></category>
            <dc:creator><![CDATA[Bartlomiej Kuchnowski]]></dc:creator>
            <pubDate>Fri, 03 Apr 2020 06:03:46 GMT</pubDate>
            <atom:updated>2020-04-03T16:05:56.137Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI in the ER: How Artificial Intelligence can transform Emergency Medicine as we know it.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TZ2zoOtUdv1uZYy_sTNCYQ.jpeg" /></figure><p>Almost everyone has been or knows someone that has been in the ER. According to the CDC, 45.8% of US citizens visited the ER in 2016 alone [1]. It is a vital part of every hospital, serving as an entry point for patients, assessing the urgency of their symptoms, conducting necessary diagnostics and giving initial treatment. That is why it is crucial for this department to operate smoothly. Yet it is not always so, especially in the time of the recent pandemic outbreak. So, how can AI contribute to alleviate some of the problems?</p><h3>Assessment</h3><p>At the entry of every Emergency Room, the patients coming in need to be triaged. It’s an initial assessment to gauge who is in need of immediate intervention and who can wait and make way for those. There are several ways and systems used by hospitals worldwide but all of them require time in order to retrieve and analyse relevant patient data. Moreover, it’s an initial assessment so every scale and algorithm has its shortcomings in identifying the outliers. That’s where Deep Learning comes in. By using available medical data, Korean researchers have created a Deep Learning network that is able to significantly outperform the existing traditional assessment algorithm (0.935 vs 0.785 AUROC).[2]</p><h3>Detection</h3><p>Sometimes, the symptoms may not be clearly visible to the human eye but are detectable for a neural network. Atrial fibrillation is a potentially dangerous condition where a patient’s arterial chambers have fast, irregular and ineffective contractions. It can be a symptom of other diseases such as pneumonia, cardiac arrest, or poisoning. A gold standard of detecting this condition is taking ECG and analysing it by a medical professional, which means it depends on equipment and staff availability. However, Chinese researchers have proposed another solution. By analysing video of the patients’ faces they were able to detect changes in the volume of the facial microvessels and detect patients who had atrial fibrillation in real time, acquiring sensitivity of 95.6% and specificity of 96.2%, respectively. [3]</p><h3>Diagnosis</h3><p>One of the most dangerous conditions is an acute intracranial hemorrhage with a 30-day mortality rate of 44%. It can be a result of a variety of conditions and events such as a head injury, hypertension, drug abuse or blood-clotting disorders. [4] A fast and accurate detection is crucial, because these patients often require life-saving neurosurgery and brain tissue is very susceptible to the lack of oxygen. A golden standard in intracranial hemorrhage detection is a computer tomography of the head (Head CAT scan), but it is difficult to do it accurately and swiftly, for a number of reasons (radiologist workload, poor signal-to-noise ratio, artifacts). It might be about to change, though, as researchers have put forward a deep learning model able to detect an intracranial hemorrhage, which is about as good as the experts from the American Board of Radiology. However, while physicians had unlimited time to examine the image, the time it took the network to do it was only one second. [5]</p><h3>Treatment</h3><p>Okay, but what about actually treating patients? Is there something AI can do to speed up the actual treatment? As it turns out, there is. Intravenous access is a standard way of delivering drugs and fluids to the hospital patients. In order to enable this, a medical professional needs to find a suitable vein (typically in the forearm), puncture the skin and get to the vessel with an over-the-needle catheter, minding that it doesn’t rupture. It takes time, expertise and sometimes several tries to do it successfully. In the ER, time is an essential asset and most drugs can not be administered before the IV access is set up. Fortunately, researchers have recently published an article in Nature Machine Intelligence journal, where they demonstrate a deep-learning robotic guidance for this. It reduced failed access attempts by sixfold and increased first-stick success rates from 53% to 88%, compared to standard manual access (gains were observed across various physiological conditions). [6]</p><h3>In conclusion</h3><p>Artificial Intelligence can significantly improve workflow, condition detection, diagnosis and treatment of ER patients. We should look forward to these advancements as these are upgrades that can super-charge our physicians, giving them more focus and time to actually save lives.</p><p>If you want to read more about AI, check out our site at Tooploox: <a href="https://www.tooploox.com/ai">https://www.tooploox.com/ai</a></p><h3>References</h3><p>[1] Rui P, Kang K, Ashman JJ. National Hospital Ambulatory Medical Care Survey: 2016 emergency department summary tables. 2016. Available from: <a href="https://www.cdc.gov/nchs/data/ahcd/nhamcs_emergency/2016_ed_web_tables.pdf">https://www.cdc.gov/nchs/data/ahcd/nhamcs_emergency/2016_ed_web_tables.pdf</a></p><p>[2] Kwon J-m, Lee Y, Lee Y, Lee S, Park H, Park J (2018) Validation of deep-learning-based triage and acuity score using a large national dataset. PLoS ONE 13(10): e0205836. <a href="https://doi.org/10.1371/journal.pone.0205836">https://doi.org/10.1371/journal.pone.0205836</a></p><p>[3] Yan BP, Lai WHS, Chan CKY, et al. High-Throughput, Contact-Free Detection of Atrial Fibrillation From Video With Deep Learning. *JAMA Cardiol.* 2020;5(1):105–107. <a href="https://doi.org/10.1001/jamacardio.2019.4004">https://doi.org/10.1001/jamacardio.2019.4004</a></p><p>[4] Intracranial Hemorrhage <a href="https://emedicine.medscape.com/article/1163977-overview">https://emedicine.medscape.com/article/1163977-overview</a></p><p>[5] Kuo W, Hӓne Ch., Pratik Mukherjee P., Malik J., Yuh EL. Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning. Proceedings of the National Academy of Sciences Nov 2019, 116 (45) 22737–22745; <a href="https://doi.org/10.1073/pnas.1908021116">https://doi.org/10.1073/pnas.1908021116</a></p><p>[6] Chen, A.I., Balter, M.L., Maguire, T.J. <em>et al.</em> Deep learning robotic guidance for autonomous vascular access. <em>Nat Mach Intell</em> <strong>2, </strong>104–115 (2020). <a href="https://doi.org/10.1038/s42256-020-0148-7">https://doi.org/10.1038/s42256-020-0148-7</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7db9c1a803d5" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/ai-in-er-how-artificial-intelligence-can-transform-emergency-medicine-as-we-know-it-7db9c1a803d5">AI in ER: How Artificial Intelligence can transform Emergency Medicine as we know it.</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Vision vs coronavirus: using deep learning augmented computer vision to help amid the current…]]></title>
            <link>https://medium.com/tooploox-ai/ai-vision-vs-coronavirus-using-deep-learning-augmented-computer-vision-to-help-amid-the-current-fe51ad7aed8e?source=rss-19daa8764b45------2</link>
            <guid isPermaLink="false">https://medium.com/p/fe51ad7aed8e</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[covid19]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[coronavirus]]></category>
            <category><![CDATA[public-health]]></category>
            <dc:creator><![CDATA[Bartlomiej Kuchnowski]]></dc:creator>
            <pubDate>Thu, 19 Mar 2020 09:50:02 GMT</pubDate>
            <atom:updated>2020-04-03T06:07:16.976Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI Vision vs coronavirus: using deep learning augmented computer vision to help amid the current health crisis.</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UPwA1ZMCK9qOPsZffC_r0g.jpeg" /></figure><p>The last few weeks have been increasingly trying to the health systems around the world as the novel coronavirus spread has been declared a pandemic by the WHO officials [1]. The sudden surge in patients needing hospitalisation and assisted respiration has put a tremendous strain on doctors, nurses and other medical staff. So is there any way we could use AI-aided computer vision to help in the current and future situations like this?</p><h3>Public screening</h3><p>Considering many people infected with the virus develop a high fever, airports and points of transit around the world have implemented manual temperature checks. Chinese tech giant, Baidu, has another idea. Using deep learning algorithms they have developed a tool for mass screening with thermographic cameras and it is currently being used at Beijing’s Qinghe Railway Station where it can examine up to 200 people at a time. All of this without disturbing natural passenger flow.</p><h3>Hand washing compliance</h3><p>Currently, one of the most important aspects of COVID-19 spread prevention is proper hand washing and disinfection. This could be helped by AI too. Researchers at Stanford PAC have shown a successful attempt at creating a computer vision system that can monitor if people walking through the entry use the hand gel dispenser or not. [3] Another study from Leibniz University Hannover proved that you can use a simple camera and a deep learning model to recognise specific hand gestures done while properly washing your hands and tell whether you did the whole routine properly. [4]</p><h3>Patient monitoring</h3><p>The last two examples were about preventive measures but what about developed cases?</p><p>As the hospitals and Intensive Care Units become more and more overcrowded, there is going to be a need for automating some of the tasks that have to be done by overworked doctors, nurses and other medical professionals. Well, there is still plenty to do. For example, there was a paper published in Nature last year about Intelligent ICU for Autonomous Patient Monitoring showing several data categories that can be gathered by a Computer Vision system such as facial expression, posture and movement analysis and sleep patterns, which can be later used to assess current patient status and his/her treatment. [5]</p><p>Another company, Oxehealth, provides cameras with software that can detect basic vital signs from a patient such as pulse rate, breathing rate or changes in activity without the need of a patient’s supervision 24/7. [6]</p><h3>Conclusions</h3><p>Apart from analysing available electronic health records, Artificial Intelligence systems can be utilised in both preventive measures and clinical care. As the situation gets more and more severe we should strive for further automation of health-related tasks in order to give medical professionals more time to do the most important part of their job: saving lives.</p><p>If you want to read more about AI, check out our site at Tooploox: <a href="https://www.tooploox.com/ai">https://www.tooploox.com/ai</a></p><h3>References</h3><p>[1] <a href="https://www.who.int/dg/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-19---11-march-2020">https://www.who.int/dg/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-19---11-march-2020</a></p><p>[2] <a href="https://www.technologyreview.com/s/615342/how-baidu-is-bringing-ai-to-the-fight-against-coronavirus/">https://www.technologyreview.com/s/615342/how-baidu-is-bringing-ai-to-the-fight-against-coronavirus/</a></p><p>[3] <a href="https://aicare.stanford.edu/projects/hand_hygiene/">https://aicare.stanford.edu/projects/hand_hygiene/</a></p><p>[4] Dietz, A., Pösch, A., Reithmeier, E. (2018): Hand hygiene monitoring based on segmentation of interacting hands with convolutional networks, Proc. SPIE 10579, Medical Imaging 2018: Imaging Informatics for Healthcare, Research, and Applications, 1057914 <a href="https://doi.org/10.1117/12.2294047">https://doi.org/10.1117/12.2294047</a></p><p>[5] Davoudi A., Malhotra K. R., Shickel B., Siegel S., Williams S., Ruppert M., Bihorac E., Ozrazgat-Baslanti T.,Tighe P. J., Bihorac A., Rashidi P (2018): The Intelligent ICU Pilot Study: Using Artificial Intelligence Technology for Autonomous Patient Monitoring. <a href="https://arxiv.org/pdf/1804.10201.pdf">https://arxiv.org/pdf/1804.10201.pdf</a></p><p>[6] <a href="https://www.oxehealth.com/the-science-behind-oxehealth">https://www.oxehealth.com/the-science-behind-oxehealth</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fe51ad7aed8e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/ai-vision-vs-coronavirus-using-deep-learning-augmented-computer-vision-to-help-amid-the-current-fe51ad7aed8e">AI Vision vs coronavirus: using deep learning augmented computer vision to help amid the current…</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Severstal Kaggle Summary — Why your team should get into data science competitions]]></title>
            <link>https://medium.com/tooploox-ai/severstal-kaggle-summary-why-your-team-should-get-into-data-science-competitions-e19eb7a84d5c?source=rss-19daa8764b45------2</link>
            <guid isPermaLink="false">https://medium.com/p/e19eb7a84d5c</guid>
            <category><![CDATA[kaggle]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Bartlomiej Kuchnowski]]></dc:creator>
            <pubDate>Thu, 19 Dec 2019 07:53:33 GMT</pubDate>
            <atom:updated>2019-12-19T13:23:42.316Z</atom:updated>
            <content:encoded><![CDATA[<h3>Severstal Kaggle Summary — Why your team should get into data science competitions</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*jeQMvJ5n4MSdK5h4.jpeg" /></figure><p>Sometimes there comes a time when there is little or no work in the office (in between the projects) and we like to spend it effectively improving our skills and broadening our horizons. Last time we had a chance to do so was in October, when we decided to assemble a MicroscopeIT team and take part in a data science competition.</p><p>We agreed to use Kaggle, a website hosting this kind of contests in various fields (e.g. object detection, natural language processing, time series analysis and prediction), and chose to compete in Severstal Kaggle Challenge, even though we were to join halfway through. It was an image segmentation challenge so it was right up our alley while the dataset and size of the competition attracted a large part of data science community, making it easier for us to learn and adapt.</p><p>For the three of us (out of five) it was an introduction to the site and first real participation in a data science competition. So here’s a short summary of the events that transpired, their outcomes and the lessons we’ve learned.</p><h3>The problem</h3><p>Our task was to detect defects in the images of steel sheets produced in Severstal factories. It was a segmentation task, so not only we would have to properly label one of the four types of defects (if there were any) but also produce a mask of pixels as close to the original annotation as possible.</p><p>Our eventual predictions would be evaluated with a DICE score.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/741/0*OEaQwGPa7ptz22bJ" /><figcaption>Examples of the annotated data. (4 different classes)</figcaption></figure><h3>What we did</h3><p>We approached the problem in a traditional fashion by familiarizing ourselves with the competition data. First of all, we split the work into 4 tasks:</p><ol><li>Research</li></ol><p>This was done by all of us in between the tasks. We researched different approaches and sifted through other public kernels in order to find the optimal solution.</p><p>2. Exploratory Data Analysis</p><p>You can’t really solve a data science problem without understanding the data you are working with. This task was done really well by Joanna and it was a huge help in terms of designing and implementing an appropriate pipeline.</p><p>3. Processing pipeline</p><p>This task was done by Marcin, Mirek, and me, with Marcin and Mirek providing initial pipelines and functions and me doing the stitching work. I was also responsible for 4), which was later improved by Marek.</p><p>4. Integrating our work.</p><p>It looks good on paper, but 3 and 4 turned out to be serious issues for most of us, mainly because we weren’t accustomed to Kaggle website and its inner workings. In hindsight, our plan was too ambitious and we spent too much of our precious time doing groundwork for the model implementation and training instead of training and trying different algorithms to solve the task.</p><p>We managed to design and implement Github-Kaggle integration because we felt Kaggle wouldn’t allow us to quickly integrate our work and keep track of the changes made by other team members.</p><p>We did a fine preprocessing pipeline and Mirek even did memory optimisation to better accommodate for the memory space Kaggle has given us. (One quirk was that on the Kaggle machines we were given 13 or 16 GB of RAM and only 5 GB of additional hard drive, so it was better to juggle the data in memory than unpack it in convenient format on hard drive).</p><p>However, all of that was too ambitious and left us with little time to train the actual algorithm. Due to Marcin’s work we managed to train a segmentation model that scored quite well, but we weren’t successful in adding the classificator trained by me, which might have let us win a medal. (the plan was to first classify an image based on whether it contained any defect and segment in another pass with a segmentation network.)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JhW_PwaRwnr1Jz9edADepQ.png" /><figcaption>MicroscopeIT team at work.</figcaption></figure><h3>The outcomes</h3><p>Our score was close to reaching a medal (14th percentile) but we fell short by 0.2 percent points in DICE score. The variance in the scores wasn’t large (0.89672 DICE score compared to 0.90883 achieved by the winners). Considering the first place method, which used classifier with a segmentator, our approach was correct but we lacked the time to fine-tune it and achieve a better result.</p><h3>Lessons learnt</h3><p>Despite the lack of apparent success on the leaderboard and the feeling of dissatisfaction we had because we felt we could have scored more if we had 2 more days, we were happy to participate and learn from the experience. We now know a little more on how to operate in Kaggle environment, how to design and train models there and what to be wary of in such time-constrained conditions. We also realised that high coding standards can actually be a hindrance, especially if you are not creating a model for a client but are on a research stage of the project.</p><h3>Conclusions</h3><p>As mentioned at the start, for most of us it was the first time we competed in a data science competition. The pressure was there, especially considering we had less than half the time compared to the rest of the participants (we joined in the last of almost three months).</p><p>We are used to doing things production-wise so even on the research stage we usually code with later usability in mind and as I’ve mentioned it isn’t always an advantage when you have little time. The good thing was that the team spirit remained high until the end and most of the frustration we experienced was connected with various Kaggle quirks we weren’t used to, rather than poor communication or tension within the team.</p><p>Also, even though some of us hadn’t had a chance to work together on a project before, the cooperation between the members was still on point, often giving the right boost at the right time, even if things seemed dire. In terms of the results and satisfaction, we collectively agreed that we were left wanting more, now that we have experience with tackling such competitions and will probably try ourselves in the future competitions, whether in-house or on our own.</p><p>In my opinion, we have learned plenty and proved that even under the pressure of time we still can have good communication and cooperation, which, to my mind, we often underestimate, since we get on well with each other on a daily basis. All these benefits we gained from the experience will definitely help us provide better results and advance further in the future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e19eb7a84d5c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/tooploox-ai/severstal-kaggle-summary-why-your-team-should-get-into-data-science-competitions-e19eb7a84d5c">Severstal Kaggle Summary — Why your team should get into data science competitions</a> was originally published in <a href="https://medium.com/tooploox-ai">Tooploox AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>