Januka Samaranayake

👉🏼 The Revival of Essay-Type Questions in Medical Education: Harnessing Artificial Intelligence and Machine Learning 🤓 Muhammad Shahid Shamim 👇🏻 https://lnkd.in/enxhcQss 🔍 Focus on data insights: - AI-driven platform ChatGPT used to grade summative essays - Innovative ways proposed for medical educators: Automated grading, content analysis, plagiarism detection, formative assessment 💡 Main outcomes and implications: - ChatGPT provided comprehensive feedback on writing skills - AI and ML technologies can enhance the assessment of essays in medical education 📚 Field significance: - Potential for AI and ML to supplement human grading in medical education - Need for further research to confirm effectiveness 🗄️: #MachineLearning #ArtificialIntelligence #Essays #ChatGPT #FormativeAssessment

1

👉🏼 ScholarGPT's performance in oral and maxillofacial surgery 🤓 Yunus Balel 👇🏻 https://lnkd.in/eb8RuGDb 🔍 Focus on data insights: - 📊 Scholar GPT achieved an average Global Quality Scale

1

How Machine Learning Algorithms Are Improving Medical Diagnosis Accuracy: Accurate medical diagnosis is critical for effective patient care, and machine learning (ML) is improving this field by enhancing diagnostic accuracy and efficiency. Here are some real-life use cases demonstrating how ML is transforming medical diagnosis: 1. Early Detection of Diseases Machine learning algorithms are significantly improving the early detection of diseases such as cancer and heart disease. For instance, ML can analyse medical imagery to detect, measure, and analyse tumors more quickly and accurately than human professionals. This capability not only speeds up the diagnostic process but also reduces referral wait times for high-risk patients, thereby saving lives. 2. Enhancing Consistency and Reducing Human Error Human specialists can be affected by factors such as fatigue and variability in interpreting medical data. ML algorithms, however, provide consistent and accurate diagnoses by identifying patterns in data that might be missed by humans. For example, ML is being used to diagnose diabetic retinopathy, age-related macular degeneration, and glaucoma with high accuracy, reducing the likelihood of diagnostic errors. 3. Broadening Healthcare Access Human specialists can be affected by factors such as fatigue and variability in interpreting medical data. ML algorithms, however, provide consistent and accurate diagnoses by identifying patterns in data that might be missed by humans. For example, ML is being used to diagnose diabetic retinopathy, age-related macular degeneration, and glaucoma with high accuracy, reducing the likelihood of diagnostic errors. 4. Improving Diagnostic Accuracy with Causal Inference Machine learning is already being used to diagnose various conditions such as certain cancers, diabetic retinopathy, Alzheimer’s disease, heart disease, and COVID-19. For example, in breast imaging, ML algorithms have achieved high diagnostic accuracy, with AUCs ranging between 0.868 and 0.909 for diagnosing breast cancer on mammograms and other imaging modalities. Real-World Impact Machine learning is already being used to diagnose various conditions such as certain cancers, diabetic retinopathy, Alzheimer’s disease, heart disease, and COVID-19. For example, in breast imaging, ML algorithms have achieved high diagnostic accuracy, with AUCs ranging between 0.868 and 0.909 for diagnosing breast cancer on mammograms and other imaging modalities. While there are challenges to widespread adoption, such as the need for more robust clinical validation and regulatory guidance, the potential benefits of ML in medical diagnostics are immense. By continuing to develop and refine these technologies, we can look forward to a future where accurate and timely medical diagnoses are accessible to all. Feel free to share your thoughts and experiences with ML in healthcare! #MachineLearning #HealthcareInnovation #AIinHealthcare #TitusAI

3

🚀🚀🚀Conferences are the best hotspot to network with your potential future collaboration partner. 👨🏻‍💻👨🏻‍💻 I am attending HAICon 2025. Come say hello if you are participating the conference today. NAAMII #healthai #healthinformatics #kathmandu #networking

15

Endless possibilities vs definitive end points. Identify the end point before looking at the endless possibilities. Wearing my practicing clinician hat, I feel the above is the biggest problem when it comes to #health #ai solutions. AI and #genai have given rise to endless possibilities (just check out the mindboggling video in Andreas Horn’s post below). When it comes to the application of these technologies in #healthcare however, it is imperative for us to realise that in the Health Industries, every single one of us is working towards a definitive end point. Before embarking on building or buying health technology solutions in this space, time needs to be spent on identifying solutions of true value and impact for there are often less than you might first imagine. Identify the end point before looking at the endless possibilities, not the other way around, at least when it comes to healthcare. HiruX Pte. Ltd. Rafael J. Grossmann, MD, MSHS, FACS Professor Shafi Ahmed

14

5 commentaires

Google just release Med-Gemini - their latest Multimodal LLM for medical domain ▶ 91.1% accuracy on MedQA (USMLE) benchmark ▶ Outperform Med-PaLM 2 by 4.6% ▶ On 7 multimodal benchmarks including NEJM Image Challenges and MMMU (health & medicine), Med-Gemini improves over GPT-4V by an average relative margin of 44.5%. I have been following the work from Vivek Natarajan and the team for a while and I have to say they are doing an incredible job in this domain More details here -> https://lnkd.in/gAnTMPdC Hope is that at some point this type of work will open up the possibility for the patients in the remote parts of the world to have access to world class doctors, as doctors/hospital will be able to use such technology remotely to increase their efficiency It is equally important to ensure that only doctors and hospitals use this technology for basic patient data gathering and analysis to improve their processes and to increase the overall efficiency, rather than allowing patients to use it for self-diagnosis, which is a significant global issue in itself _____________________________________ Follow me for regular updates and insights on Generative AI and my journey as a CTO in AI product development

25

2 commentaires

AI in healthcare by Adjiedj Bakas | What to expect ? interesting read ! AI in healthcare by Adjiedj Bakas Putiton-e is the paid encyclopedia Want your stories on Putiton-e.com | https://lnkd.in/exn_WHW7

5

Google just announced Med-Gemini, an advanced Al fine-tuned for medical tasks and it beats GPT-4 on all benchmarks. This new model surpasses GPT-4 across various medical benchmarks, achieving an impressive 91.1% accuracy on the MedQA-USMLE benchmark and outperforming GPT-4 by an average of 44.5% on seven multimodal benchmarks. It excels in tasks like medical summarization, generating doctor referrals, and simplifying medical documents, proving more effective than human experts for complex text-based medical tasks. This is a huge advancement in Al for healthcare, indicating potential enhancements in medical diagnostics and patient care, and positioning healthcare as a forefront of Al development with significant impacts on clinical practices and education.

8

Update on our FYP: Early Retirement: A Game with Live NPC Interaction Using LLMs 🎮🤖 Just wrapped up a project on Multilingual Sentiment Analysis for Urdu, Arabic, and English review datasets 🌍. It was part of the Ablation study, so there was a lot of trial and error 🔄, along with quite a few tweaks along the way 🔧. But in the end, I managed to hit some impressive accuracy rates: 97% for Urdu, 97% for English, and 98% for Arabic 🎯. There was a frustrating stretch where the accuracy wouldn’t go beyond 80% 😤, so I tackled it by combining an Attention Layer with a Transformer Layer and some standard architecture, which really paid off 💡. Now, I’m back to working on the chatbot, which will either integrate with WhatsApp or Telegram for a human cloning project 🤖📱. Let’s see where that goes! 🚀 In the meantime, stay tuned and keep supporting me as we push forward with this FYP 💪. #FYP #AI #LearningJourney #EarlyRetirement #FinalYearProject #LLM #Support #GameDevelopment 🎮📚

4

1 commentaire

Patient care using AI and enhancing operational efficiency

1

Check out this 2nd post from Adil Ahmed in his series about LLMs. In it you'll see how we fine tuned ClinicalBERT to work with PHI without compromising confidentiality. Adil even included a link to our github so you can see how the model works. https://lnkd.in/dx-63f9Z #HealthcareInnovation #PatientPrivacy #NLP #MachineLearning #HealthcareAI

3

2 commentaires

✨Excited to share my latest project : Cura AI - Your Personalised Healthcare Assistant Cura AI - a mobile app that will transform healthcare with Agentic AI. Unlike traditional AI, Agentic AI guides users. Offering intelligent system to analyse symptoms , personalise treatment plans and offer continuous health monitoring. ✨It’s like having a proactive health companion in your pocket Check out the detailed case study to see how Cura AI will be enhancing lives with technology: https://lnkd.in/gkNNcdd4 #AI #ArtificialIntelligence #MachineLearning #DeepLearning #TechInnovation #DigitalTransformation #FutureOfTech #Innovation #Healthcare #HealthTech #DigitalHealth #MedTech #HealthcareInnovation #HealthCareAI #HealthIT #Telemedicine #eHealth #UXDesign #UIDesign #ProductDesign #UserExperience #DesignThinking #DesignInnovation #MobileAppDesign #HealthCareRevolution #SmartHealthcare #HealthAndWellness #DigitalHealthRevolution #Innovation #Startups #FutureOfWork #TechForGood #Entrepreneurship #Leadership #Growth #Impact

55

8 commentaires

As #FHIR gains popularity, we see many naive implementations destined to break once in production. By naive, I mean developers having optimistic expectations about FHIR data, often due to working with synthetic data sets and not reading the spec or being exposed to real-world data. Developers tend to design solutions around examples instead of reading specs, and this is understandable. However, to be able to communicate healthcare information, FHIR's highly nested structure is crucial and cannot be overly simplified. Yet, developers often flatten it without realizing they are actually making business decisions that carry future implications. Testing apps against simple test data makes everything seem fine. If a Patient resource has only one identifier, address, and HumanName entry (common in synthetic data), displaying this data to a user is easy. But real-world data may use the full structure of the resource: multiple addresses (home, work, different validity periods), identifiers, and HumanName entries (official, nicknames, different languages). What happens next is the result of many small, unintentional decisions developers made. Typically, one of the following occurs: 1. The app crashes. 2. Arbitrary values are displayed, often irrelevant or unsuitable (like the first entry in an array). 3. The app displays the full lists, overwhelming users with competing values, without helping them differentiate between them. Data may be truncated if it doesn’t fit the screen. Worse problems arise when the app tries to act on this data, like saving it in a non-FHIR database, passing it to another system, or triggering decision support mechanisms. If an important message is sent to an old phone number or the wrong address, there could be real-world consequences. When developers assumed these arrays only contain one element or that the first one is the most important, they unknowingly created a bug that might surface when it's too late. Of course this kind of flattening issues can occur with all FHIR resources, the Patient is just a simple example. Think about what can happen when actual clinical data is treated with the same naivety. This brings us to another often-overlooked aspect: #QA. If your QA team has a solid understanding of FHIR and common developer mistakes, they can create better test data that exposes these issues before production. But that’s a topic for another post... So how to properly flatten FHIR data? Look at the spec, see if an element can repeat and what attributes it may have that can help differentiate between the entries. Define clear rules for how to extract the most relevant entry from each array. This may depend on context - one app may need to select the home address that was valid at the time of an Encounter, another app may prefer the most current billing address. Just a few more lines of code, and you have a more robust FHIR application. #HealthTech #DataIntegrity #QualityAssurance #HealthcareData #ClinicalData

45

7 commentaires

I continued to scratch an architectural itch and added more features to generative_redfoot (see https://lnkd.in/gB26jt99 ). It can now: - Take values for variables to use as the initial context for the PDL program - Use mlx cache for model generation if indicated in PDL - Provide 'named' sections of text from a Wordloom file to use in building a prompt in a language-agnostic way. - Provide few shot examples from a file to use in building/loading a CoT prompt Finally, You can specify Google's google/gemma-7b-aps-it model to perform "abstractive proposition segmentation" from LLM output, etc., and either print it or use it for subsequent PDL logic So far, I like how it neatly separates concerns between prompt "language" management, construction, and LLM "workflow" management/orchestration. The expressive power of PDL is very useful for this. It should be easy to extend with HF's transformers, CUDA libraries, and other non-Apple Metal ecosystems (I just happen to use the Apple ecosystem for the very active open-source community and more cost-effective hardware).

3