Initiate Ventures’ Post

Meet Michael Elashoff, founder and Co-CEO of Cornerstone AI. Mike has a strong background in biostatistics and data science, with degrees from UC Berkeley and Harvard. Prior to founding Cornerstone, Mike held positions with the FDA, Medidata, and Project Ronin. He also founded Patient Profiles (acquired by Medidata). At every step of his career, he's encountered the same issue time and again—the very manual, very time consuming, and very necessary task of cleaning data. This problem begged a question: what if this was a problem that could be solved? Learn more about Mike and the genesis of Cornerstone: https://lnkd.in/gfzGQ7gt #MeetTheTeam #DataScience #BioStatistics #DataCleaning #AI

🌟 Bridging Data Science and Clinical Research: A Statistician’s Perspective 🌟 As a statistician passionate about Data Science and Clinical Research, I find myself at the intersection of two worlds—both driven by data and innovation. 📊 In Clinical Research, data plays a critical role in transforming patient care. From trial design to outcome analysis, statistics empower researchers to make evidence-based decisions that save lives. By leveraging machine learning and predictive analytics, we can uncover patterns and trends that were previously hidden, making healthcare more efficient and effective. 🔍 What I'm Focused On: Statistical Analysis for clinical trials and drug development, Data Science Tools to optimize research outcomes, AI & Machine Learning to predict treatment efficacy and patient outcomes. Being a part of this revolution in both data science and healthcare is truly exciting, and I’m committed to exploring how data can continue to drive breakthroughs in medical science! 🚀 #DataScience #ClinicalResearch #Statistics #AI #HealthcareInnovation #BigData #MachineLearning

A mistake to avoid in Clinical Research, Biostatistics and Data Science. Analysis without considering the subgroups can be a big mistake. Very often, it can lead to Simson's paradox (also called the Yule–Simpson effect) in which the overall data studied may show trends which actually diminish or even be completely reverse when subgroups are evaluated. What most don't really consider is that missing out on subgrouping may also lead to causal analysis failure as well and latent variable structure having more bias. This effect may be countered using different strategies from mainstream statistical ones to machine learning based ones of identifying and making the subgroups in the data : 1. Domain specific knowledge and subgroup labeling adding the biological context and identifying 2. Unsupervised Clustering as a type of unsupervised analysis used to identify data based subgroups 3. Any confounder correction principle will help, of course the randomization will help but other approaches may still be needed. 4. CART - classification and regression trees are one of my favorite aproaches in data driven subgroup identification. 5. Feature engineering based on both biological domains and data driven aproaches. 6. Last in this post but very important. AI aided subgorup analysis. AI models can be trained to identify the subgroups based on the biological labeling and the covariates of interest and using as a modeling augmentation aproach. You wont hear much about this one as its still a developing field. #datascience #biostatistics #clinicalresearch #analytics #ai #artificialintelligence #machinelearning #biology #ml #stats #biomedical

In the ever-evolving landscape of healthcare, the quest for deeper insights and more precise interventions remains paramount. Amidst this pursuit, unsupervised learning emerges as a powerful tool, offering the promise of uncovering hidden structures within data. Unlike supervised methods that target specific outcomes, unsupervised learning operates in a more exploratory fashion, aiming to reveal underlying patterns and relationships. In this article, we delve into the realm of unsupervised learning within the healthcare domain, exploring its applications, challenges, and potential impact. Join us at the Clinical Operations and Clinical Trials Supply Chain Strategy Meetings in Princeton and San Diego to delve deeper! Register for Princeton - https://lnkd.in/d_TWpq98 Register for San Diego- https://lnkd.in/dMys-6zb #unsupersivedlearning #machinelearning #clustering #electronicmedicalrecord #multiclassmodels #predictivepathways #clinicaldatasources #datadimensionality #clinicaltrials #clinicaloperations

In healthcare's dynamic landscape, the pursuit of deeper insights and precision interventions is paramount, where unsupervised learning emerges as a potent tool for revealing hidden data structures. https://lnkd.in/gAZHG4sr #unsupersivedlearning #machinelearning #clustering #electronicmedicalrecord #multiclassmodels #predictivepathways #clinicaldatasources #datadimensionality

👁🗨 Exciting Journey in Cataract Disease Detection with Data Science! 🚀 🔍 Dataset Exploration: Immersed myself in Kaggle's Cataract Disease Dataset, a treasure trove of insights and challenges in medical image analysis. 🧠 Data Science Alchemy: Wielded a custom Convolutional Neural Network (CNN) to decipher intricate patterns within medical images, a pivotal tool in the fight against Cataract disease. 🔄 Imbalance Conundrum: Encountered the pressing issue of class imbalances. The balance between too little and too much augmentation became the fulcrum of success. Augmenting all classes equitably emerged as the silver bullet, countering biases and fostering a resilient model. 📊 Navigating Validation: The validation tightrope – finding equilibrium in augmentation intensity. Too little risked biases, while excessive augmentation inflated validation loss. Striking the perfect balance emerged as the cornerstone for a robust model. 🛠 Model Triumph: Overcame hurdles, achieving a validation accuracy exceeding 90%. The model now stands poised for precise predictions of Cataract disease. 🌐 Global Accessibility: Seamlessly deployed the model using Ngrok, transcending geographical barriers and making healthcare technology universally accessible through a public link. 🔄 Iterative Excellence: The iterative loop of training, testing, and refinement sculpted a model capable of reliable disease detection, a testament to persistence and innovation. 💡 Key Learnings: Collaboration proved indispensable. A friend sought assistance, and together, we triumphed over real-world dataset challenges. 🌟 **Success Shared:** This project exemplifies the synergy of data science and healthcare, solving tangible issues and delivering meaningful impact. 🌐💙 🤔 Expert Insights Sought: What are your go-to strategies when faced with extensive class imbalances? How do you address these challenges, and what solutions have proven most effective in your data science journey? #CataractDisease #DataScience #KaggleDataset #CustomCNN #AI #machinelearning #computervision #kaggle #flask #DataAugmentation #NgrokDeployment #HealthTech #MachineLearning #CollaborationSuccess #MedicalImageAnalysis #InnovationInHealthcare

🔍 𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗻𝗴 𝗖𝗵𝗿𝗼𝗻𝗶𝗰 𝗞𝗶𝗱𝗻𝗲𝘆 𝗗𝗶𝘀𝗲𝗮𝘀𝗲 (𝗖𝗞𝗗) 𝘄𝗶𝘁𝗵 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴: 𝗔 𝗖𝗼𝗺𝗽𝗮𝗿𝗮𝘁𝗶𝘃𝗲 𝗦𝘁𝘂𝗱𝘆 In the ever-evolving field of healthcare, leveraging machine learning for early detection of chronic conditions can be a game-changer. Recently, I conducted a study that explores the prediction of Chronic Kidney Disease (CKD) using two different machine learning algorithms: Linear Regression and K-Nearest Neighbors (KNN). 🧠 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: 𝙊𝙗𝙟𝙚𝙘𝙩𝙞𝙫𝙚: The goal was to predict CKD in patients based on their medical test results using machine learning models. 𝘿𝙖𝙩𝙖: The dataset comprised 158 patient records with 26 features, including vital signs, laboratory test results, and clinical observations. 𝙈𝙤𝙙𝙚𝙡𝙨 𝙐𝙨𝙚𝙙: Linear Regression and KNN were employed to develop predictive models, which were then compared based on their accuracy and other metrics. 🔗 𝗥𝗲𝘀𝘂𝗹𝘁𝘀: Linear Regression achieved an accuracy of 80%. Its simplicity and high interpretability make it a solid choice for modeling, but it is sensitive to outliers and assumes a linear relationship between variables. K-Nearest Neighbors (KNN) outperformed Linear Regression with an accuracy of 85%. KNN’s non-parametric nature allowed it to capture complex, non-linear relationships in the data, making it more suitable for this particular task. 🎯 𝗖𝗼𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻: The study concluded that KNN is more effective in predicting CKD due to its ability to adapt to the complex patterns within the medical data. This finding underscores the importance of selecting the right algorithm based on the specific characteristics of the dataset at hand. #MachineLearning #HealthcareAI #DataScience #ArtificialIntelligence #HealthTech #Innovation #PredictiveModeling Boston Institute of Analytics AWS Machine Learning

🖋 The study finds that the clinical validity of the SMOTE technique for addressing class imbalance in medical datasets is questionable due to its inability to effectively predict minority-class outcomes, unlike its performance in financial datasets, highlighting the need for advanced and clinically relevant modeling techniques. #medicaldata; #imbalanceddataset; #minorityclassprediction; #machinelearning; #deeplearning; #cost-sensitivelearning; #ensemblelearning; #SHapleyAdditiveexPlanations(SHAP); #featureimportance; #principalcomponentanalysis (PCA) "Impact of Nature of Medical Data on Machine and Deep Learning for Imbalanced Datasets: Clinical Validity of SMOTE Is Questionable" By Seifollah (Seif) Gholampour 🔗 https://lnkd.in/gfhpxnsh

Title: "COVID-19 and Beyond: Enhancing Pneumonia Detection Through Advanced Deep Learning and Multi-Modal Data Integration" I am thrilled to share the culmination of months of dedicated work on my latest project! Introduction: In the relentless battle against pneumonia, particularly heightened by the COVID-19 pandemic, I am excited to present my dissertation, "Advancing Pneumonia Detection in the Era of COVID-19." This research dives into innovative methodologies, leveraging cutting-edge deep learning techniques and multi-modal data integration. Background: Over the past challenging years, COVID-19 has reshaped our world, impacting every sector of society. With over 200 million confirmed cases and 4 million deaths, the urgency for enhanced healthcare solutions is undeniable. This project addresses the critical need for advanced pneumonia detection, especially within the context of the ongoing pandemic. Motivation: The motivation behind this study is rooted in the imperative for precise and timely pneumonia detection. Beyond the immediate challenges posed by COVID-19, early identification plays a pivotal role in improving patient outcomes, reducing healthcare system burden, and contributing to the management of respiratory diseases on a global scale. Objectives: The primary objectives include the development of an advanced deep-learning model for pneumonia detection and the integration of multi-modal data sources to improve diagnostic accuracy. The Model: I am thrilled to share that the heart of this research is a powerful deep-learning model built on TensorFlow. Leveraging the VGG16 architecture from TensorFlow's Keras applications, this model underwent rigorous training and evaluation. With the use of ImageDataGenerator for robust data augmentation, the model achieved an accuracy of 0.9167 during testing, showcasing the robustness of our model in accurately detecting pneumonia. This hands-on experience underscores the commitment to innovation and precision in the pursuit of better healthcare solutions. Conclusion: As we navigate these unprecedented times, the convergence of public health, technology, and society demands adaptive and comprehensive solutions. This project endeavors to make a meaningful contribution to the ongoing efforts to address complex challenges brought forth by the intersection of a global pandemic and healthcare disparities. I am excited to share the journey and eager to hear your thoughts! Let's drive innovation together. #HealthTech #DeepLearning #COVID19Research #PneumoniaDetection

I am impressed by this article about the new Chief Data Scientist of Duke Health, Michael Pencina. He is a pioneer in AI and data science, and he has a bold vision for making healthcare more equitable, personalized, and accessible.   As a healthcare enthusiast, I appreciate his work and his goals. I am curious to see how he and his team will use AI and data science to improve health outcomes and quality of care for patients and communities.   What are your thoughts on this article? How do you think AI and data science can impact healthcare? Let me know in the comments below. #Healthcare #AI #DataScience #Innovation #Equity #Duke #DukeAI #MicrosoftHealth #IBMHealth #GoogleHealth #AppleHealth #IBMWatson #healthequity https://lnkd.in/eFqg39Mm