Week 4 - Juan Boza: A whole lot of Motivation!

A whole lot of Motivation!  

Week 4 has ended... Wow... What a fantastic four weeks, honestly. I cannot explain how much I have learned these past weeks - about the clinic, the research, and myself. Believe it or not, being exposed to the clinical difficulties, patient stories, and medical procedures have changed me in many ways. While some might prefer to avoid these exposures, I see them as MOTIVATION. Because of them, I am more motivated and determined to develop better diagnostics. Let me explain a little more - First, it is easy to get overwhelmed with all the devices present in the clinical setting. Thus, the development of compact and easy-to-use devices is crucial for its successful implementation. Second, every patient has a story worth hearing and understanding, especially when they have been looking for an answer to their situation for a very long time. Third, certain diagnostic procedures can be traumatic for the patient and yet not produce an answer. For this, further technology should be developed to reduce such procedures when unnecessary. 

  Fig 1. Nail psoriasis. Common characteristics include

  pitting and onycholysis. Sometimes inflammation 

  is present. Patients feel pain and tenderness.

Dermatology huh? Are you ready for this week's nail condition edition? Today I will teach you about psoriasis. Did I just say psoriasis? you bet I did... This might be a shock for you, but you can actually have psoriasis in your nails, and only in your nails for some people. Whaaaaaaat???? well, yes, some people do present with nail psoriasis even if the common rashes behind the neck or the elbows are absent. I bet you are wondering how does it look like, so take a look at Fig 1. Here you can observe the most common aspects of this autoimmune disease - Nail pitting (dentures on the nail), inflammation, tenderness, and onycholysis (separation of the nail from the nail plate). Just so you know, this is a painful condition and has no cure, so treatment is performed every so often depending on the case. Commonly, nail psoriasis can be confused with onychomycosis, which is a fungal infection that we will cover in next week's special. 

Regarding other activities, research was very fruitful this week. I learned about AI implementation for image classification. For this, I coded more than 10 different AI models and implemented them with a Dogs vs. Cats dataset. This was done to make sure that the AI is well set up and ready for implementation. An important aspect is that I coded these AIs from scratch, which requires large datasets for proper training. Said this, I decided to use pre-trained versions of these models and retrain them for nail diagnostics. The plan consists of training all these models and determining which of them will work best for this specific application. Fig 2 shows the results of transfer learning the ResNet152 model. The datasets are comprised of 600 images split in half for training and testing (150 dogs, 150 cats each). The training/testing process for each epoch (iteration of training with the same data) is depicted in Fig 2.A. The accuracies for both reach >95%, proving that the model can be used for differentiating between dogs and cats. Fig 2.B shows the images along with the predicted classes from the testing batch. 

    Fig 2. A. Accuracy of the AI model ResNet-152 after transfer learning for Cats Vs.
    Dogs. The accuracy is plotted against each epoch. B. Prediction of test images for
    images (small subset).

As you can see, I had a very busy week. Next week, I hope to be able to use actual nail images to train and test the AIs. I hope your week was as good as mine. 

Stay tuned...