Natural Language Processing contributes to Suicide Prevention and Breast Cancer Diagnosis

The rapid growth of NLP

Natural word processing (NLP) is based on the understanding, manipulation and generation of natural language by machines. It allows machine translation, sentiment analysis, chatbots, text classification, speech recognition, optical character recognition (OCR), automatic correction, automatic summary generation, text to speech.

Since 2019, thanks the Artificial Intelligence departments of the GAFA, the NLP has experienced a meteoric rise with the availability of increasingly powerful algorithms such as ELmo, Open AI GPT, BERT (Google), LASER (Facebook),T5 (Google).

ELMo, GPT, BERT, T5 — Question Answering | Coursera

Natural Language Processing (NLP) is already used in the field of medicine for the prevention and diagnosis of diseases.

NLP used to prevent suicide

The World Health Organization (WHO) records 16 million suicide attempts per year worldwide. More than 800,000 people a year die by committing suicide. With approximately 9,000 deaths by suicide per year, France has one of the highest suicide rates in Europe.

In France in 2019, suicide will account for 16% of deaths between 15 and 24 years old and 20% among 25–34 year olds. The omnipresence of social media in the lives of teenagers and adults offers new types of data to understand the behavior of those who attempt to take their own lives and suggests new possibilities for suicide prevention.
The NLP makes it possible to create algorithms capable of classifying whether a text published on social networks contains suicidal allusions or not.

Building a suicidal tweet classifier using NLP

Once trained, the model can predict the depressive tone of a text.

The application of natural language processing for suicide prevention has resulted in the publication of several research papers.
- Identification of Suicide Ideas and Suicide Attempts in a Psychiatric Clinical Research Database Using Natural Language Processing (2018)
- Identifying suicidal adolescents from mental health records using natural language processing (2019)

As early as 2017, Facebook has developed an algorithm capable of detecting suicidal among its users coupled with an alert system that warns friends of identified users.

Facebook rolls out AI to detect suicidal posts before they’re reported

Alert sent when a suicidal profile is identified

The NLP for diagnosing cancer

Atypical ductal hyperplasia is an increase in the number of abnormal cells in the mammary ducts. Having atypical ductal hyperplasia increases the risk of breast cancer.

Researchers collected 24,881 breast pathology reports from seven different hospitals and manually annotated them with nine key attributes that describe the types of atypicals and cancers. They then transformed this text into “word embeddings”.

Each word was represented by a real number vector. These word embeddings are the inputs of a convolutional neural network used to extract the absence or presence of atypical ductal hyperplasia.

The model trained with data from four different hospitals reached an accuracy of 96%. By using data from different hospitals to train the model, the researchers improved the generalizability of their model to new data.

Do Neural Information Extraction Algorithms Generalize Across Institutions?

This article was written based on these resources…

MIT 6.S897 Machine Learning for Healthcare, Spring 2020- Lesson 7 & 8 NLP