Skip to main content
7 answers
12
Asked 1922 views

Where do I start off as a medical student interested in AI/ML?

I'm a medical student who's deeply interested in AI and machine learning and building valuable AI systems in healthcare and medicine. I'm quite good at Python and I've been trying to take AI/ML courses on my own, but I've never been so confused.

Where do I start? How do I proceed? Do I need all the math? All the resources I lay my hands on end up confusing me more the previous.

How do I get to the point where I'll actually start building stuff? I feel like I've been so stuck for months now.


12

7 answers


1
Updated
Share a link to this answer
Share a link to this answer

Jezlea’s Answer

The most direct path forward for someone with Python already in hand is to start with fast.ai rather than traditional ML courses. It is specifically designed to get practitioners building real things immediately and introduces theory only as it becomes relevant. From there, focus on healthcare specific applications right away rather than generic ML problems. Medical imaging classification, clinical note processing with natural language processing, and patient outcome prediction are all areas with existing open datasets and active communities where your medical knowledge is genuinely an asset that pure ML engineers lack.

The combination you are building, clinical expertise plus ML capability, is genuinely one of the most valuable profiles in all of healthcare right now and the field desperately needs people who understand both sides. Stop trying to become a machine learning engineer and start positioning yourself as a clinician who builds intelligent healthcare tools, because that framing will guide you toward the right depth of knowledge in each area without requiring you to master everything. Pick one small, focused project in an area of medicine you already know well, constrain it ruthlessly, and build something imperfect but real. That first completed project will teach you more than another six months of courses ever could.
Thank you comment icon Thank you so much for the advice. Victor
Thank you comment icon Thank you so much for taking the time to answer my question. Your advice on picking one small, focused project in an area of medicine I already know well, was super insightful and has really given me clarity on my next steps. Thanks again for the value you share! Victor
1
0
Updated
Share a link to this answer
Share a link to this answer

Priyanshi’s Answer

Hey Victor,
Honestly, you’re probably confused because you’re trying to learn *all of AI/ML* before building anything. You don’t need to. Pick one small healthcare problem, start building, and learn along the way.

For example, explore how ML models can help detect Parkinson’s disease or breast cancer. Begin with a simple public tabular dataset instead of medical images. Try logistic regression, decision trees, random forests, and XGBoost, then compare their results and understand where they fail.

You don’t need to master all the maths or code every algorithm from scratch. Focus on understanding the basic maths and logic behind each model, how it learns, what assumptions it makes, and why it gives certain predictions.

You can use libraries that already provide ML models:

- Scikit-learn: Best for beginners and traditional models such as logistic regression, decision trees, random forests, and SVMs.
- XGBoost: Great for structured or tabular medical data.
- TensorFlow/Keras: Useful for neural networks and deep learning.
- PyTorch: Commonly used for deep learning and medical-AI research.

Learn basic functions such as `fit()`, `predict()`, and `predict_proba()`, along with simple hyperparameter-tuning tools like `GridSearchCV`. You only need basic statistics, probability, vectors, and an intuitive understanding of gradient descent to get started.

Since you’re a medical student, your clinical knowledge is a major advantage. Don’t judge a model only by its accuracy. Also ask: How many cases does it miss? How many false alarms does it create? Is the data biased? Would the model work in another hospital?

For your first project:

1. Pick one disease and a clear clinical question.
2. Find a public, de-identified dataset.
3. Clean and explore the data.
4. Train a basic model using scikit-learn.
5. Compare and tune two or three models.
6. Check sensitivity, specificity, precision, recall, ROC-AUC, and calibration.
7. Analyse the mistakes from a clinical perspective.

You could also write a small literature review titled “How AI/ML models can help with the early detection of disease in healthcare.” Narrow it to Parkinson’s or breast cancer, review 10–15 papers, and compare their models, datasets, results, limitations, and clinical usefulness.

Most importantly, don’t wait until you feel completely ready. Give yourself one month to complete a small, imperfect project, it will teach you far more than taking another five courses.
0
0
Updated
Share a link to this answer
Share a link to this answer

Sandeep’s Answer

Hello Victor,

Since you already know Python, start with the basics of machine learning using scikit-learn. Learn data preprocessing, classification, regression, and model evaluation, then build a small healthcare project, such as predicting diabetes risk using a public dataset.

You don’t need to master all the math before starting. Learn the necessary statistics, probability, and linear algebra as you go. Avoid jumping between too many courses—choose one beginner-friendly resource and finish it by building projects.
0
0
Updated
Share a link to this answer
Share a link to this answer

Supreeti’s Answer

You do not need to learn all the math or finish a long list of courses before building. Since you already know Python and have medical expertise, your biggest advantage is combining those two skills. Learn ML fundamentals — supervised/unsupervised learning, train/validation/test sets, overfitting, evaluation metrics, feature engineering, and basic model interpretation. As far as math probability, statistics, linear algebra, and a little calculus per the project needs, you can learn as you apply. My recommendation would be to start building and creating a portfolio. Pick small healthcare datasets and build projects using pandas, scikit-learn, and later PyTorch. Then focus on clinical decision-support and healthcare workflows as you progress. A few well-designed projects demonstrating clinical reasoning, data handling, model evaluation, limitations, and patient-safety considerations will be far more valuable than collecting certificates. Good luck and Hope this helps.
Thank you comment icon Thank you so much for taking the time to answer my question. Your recommendation to start building and creating a portfolio was super insightful and has really given me clarity on my next steps. Thanks again for the value you share! Best regards. Victor
0
0
Updated
Share a link to this answer
Share a link to this answer

Chinyere’s Answer

Hi Victor,

You are probably stuck because you are trying to learn "AI/ML" as one huge subject. It isn't. There are several layers to it, and you don't need to master all of them before you start building.

Since you already know Python, I would actually stop collecting courses for a while and start making small projects. You will learn the theory much faster when you have a problem in front of you that you are trying to solve.

I suggest you use this progression:

1. Learn the basic data tools: Get comfortable with NumPy, pandas, matplotlib, and scikit-learn. Learn how to clean a dataset, explore it, split it into training and testing data, train a simple model, evaluate it, and explain what the results mean.

2. Learn only the math you need at each stage: You do not need to complete an entire mathematics degree before touching machine learning. Start with basic statistics and probability, then learn the relevant concepts from linear algebra and calculus as they appear in your ML work. Mean, variance, probability distributions, correlation, vectors, matrices, derivatives, and optimization are good concepts to gradually understand.

3. Start with healthcare datasets: This is where your medical background becomes a major advantage. Instead of building another generic house-price predictor, try a small project using a public healthcare dataset. For example, you could build a model that predicts a clearly defined outcome, classify medical images using an appropriate beginner dataset, or analyze factors associated with a health outcome.

4. Learn how to evaluate healthcare models responsibly: Accuracy alone is not enough in medicine. Learn about sensitivity, specificity, precision, recall, ROC-AUC, calibration, class imbalance, bias, data leakage, privacy, and external validation. A model that performs well in a notebook may perform very differently with real patients.

5. Build a portfolio as you learn: For every project, put the code on GitHub and write a short explanation of the clinical problem, dataset, approach, results, limitations, and what you would improve. Three thoughtful projects are more useful than twenty unfinished courses.

Your medical training also gives you something many technical people don't have: an understanding of clinical problems and how healthcare actually works. That combination can eventually lead toward clinical AI, medical imaging, clinical decision support, health informatics, biomedical data science, or AI research.

So you should probably change your goal from "I need to learn everything about AI/ML" to "I will build one small healthcare ML project this month." Once you build the first one, the next topics you need to learn will become much easier to identify.

You don't need to wait until you feel ready. Start small, then let your projects tell you what to learn next.

Best wishes!
0
0
Updated
Share a link to this answer
Share a link to this answer

David’s Answer

Hi Victor,

I have been in your shoes. I would actually say that you are in the perfect position to learn more about AI/ML, since you already have a specialization in a different field.

I will repeat what some other users have suggested in the forum here, but projects are your best bet. Saying I want to learn everything about AI/ML is like saying I want to learn how to paint. What brushes are you going to use? What style are you going to paint in? What canvas? What colors? What paint? Are you going to use paint, or oil, etc.?

It brings more questions than answers-- I would suggest picking the project before you pick what you want to learn. Like picking what you are going to draw/paint first, and then figuring out what materials to use afterwards. The materials will be tailored to the use case, just like the AI/ML method will be tailored towards the specific project that you will be working on.

Since you are comfortable with Python, I would suggest the following:
1. Getting comfortable with Pandas. This will allow you to curate the data that you choose, and get it ready for the training + testing cycle. For data, I recommend taking a look at Kaggle. It is an open source website that has data available, at different complexity levels. When you are getting started, I would recommend only messing with datasets rated at 8+. If that doesn't work for you, try to find other open source datasets-- universities often publish these.
2. Getting comfortable with scikit-learn. Stick to the basics at first. Regression + Classification. No complex math, just the WHY. When you would use one use case over the other. There is a lot more that you can dive into, like supervised vs. unsupervised learning, random forest, linear, clustering, reduction(PCA, etc.) and more. I recommend diving into the User Guide on the official scikit-learn page, once you get comfortable with the basics.
3. Once you have your model built out, I recommend looking into simple serving or visualization techniques. It depends on your project, but the question to ask is: will users be calling this model, or will users be shown the results of this model? If it is calling, then you need to implement some sort of API call or method that can easily call the model. I recommend FastAPI if you are just getting started, it is really intuitive, easy to learn, and comes with helpful utilities to get off the ground, issue-free. For viewing the results, then you would need to put together graphs or plots showing the results of the testing, validation, training, and final results. There are many methods to this-- seaborn, plotly, etc. Pick the one that looks the coolest-- you will be more inspired to learn it.

I think overall, the math is a tool to go deeper. It is not an entry level requirement to get into AI/ML. It's like Chemistry for being a Med Student. You don't need it to learn anatomy, but it will definitely help you take your understanding to a deeper level.

For now, focus on different projects that you will enjoy, and come back to. That's what continues to work for me.

Good luck, and enjoy the ride!
0
0
Updated
Share a link to this answer
Share a link to this answer

Karolina’s Answer

Getting started in AI can feel overwhelming because there are thousands of guides, docs, and videos thrown at you all at once. The good news is that almost everything you need to learn is out there. The tricky part is just figuring out how to navigate it.

When I was first introduced to Claude, I was overwhelmed by all of the resources available. So I took a step back and asked myself, " What am I actually trying to learn/do?" For me, it was just the basics:
- What does this tool actually do?
- How do I install it?
- How can I try it out myself?

I went on YouTube and searched "Claude Code for beginners" and found a video titled "Claude Code for Absolute Beginners: Step by Step Tutorial" and followed along. Watching someone use the tool step by step gave me the foundation I needed to start experimenting on my own. Today, I'm creating AI agents and automations that save me hours every single day.

So, ask yourself, what are you actually trying to learn or do (the basics, how to build a specific workflow to help with xyz, etc)? The key takeaway: You don't need to understand everything on day one. You can find a simple step by step video and follow along, or talk to the AI tool to help you build what you're trying to do.

Karolina recommends the following next steps:

Ask yourself, what are you actually trying to learn/ do/ build (be specific)?
Find your modem of learning (visual learner --> videos, reading --> forums, interactive --> talking to an AI tool/LLM) and use the tool to help you get started (Youtube, AI podcasts, Chatgpt/Claude/Gemini/etc).).
Start browsing the content with your curated results. Rephrase if you don't get what you need right away. Or if using an AI tool/ LLM, talk to the tool, ask for clarity when needed.
0