Multiple researchers at Effat University are doing their part to push forward the ability of AI to spot skin cancer early and save lives, including a new method to analyse possible breast cancers.
Early detection is essential in the fight against cancer. Skin cancer has an alarmingly high mortality rate, but up to 95% of people can survive cancer as long as it is treated soon enough.
It takes a specialise to know whether a tumour on the skin is benign or malignant, and those specialises are in short supply. The availability of skilled dermatologists and the time required for comprehensive analysis can bottleneck cancer diagnosis, particularly in regions with a high patient load or limited access to medical facilities.
The solution: artificial intelligence
Computers programmed to recognise patterns and trained on a library of tumour images can look at a new picture of a patient's tumour and sort it into the correct category. This is called machine learning.
There are many models for machine learning and it can be a complex field. A recent paper co-authored by Effat University’s Saeed Mian Qaisar, identifies and compares various techniques. This included the highly accurate Support Vector Machines (SVMs) (developed in the 1990s) and the flexible K-means Clustering and K-nearest Neighbours (KNNs) (developed in the 1960s).
The highest levels of accuracy are found with deep learning models, of which there are several, including long short-term memory (LSTM), Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs). The latter of which accurately predict different types of skin cancer with over 90% accuracy.
For the full comparison of 17 machine learning and deep learning techniques, read the full paper in our research repository:
AI-Powered Diagnosis of Skin Cancer: A Contemporary Review, Open Challenges and Future Research Directions.
The future of AI medical diagnosis
This research indicates that CNNs are the best tool we currently have in skin cancer diagnosis, but this area is progressing all the time. One paper co-authored by Effat University Abdulhamit Subasi, proposes the “grid-based deep feature generator”, which divides an image of suspected breast cancer into rows and columns and applies pre-trained CNN models to each row and column.
You can find the full study here:
Artificial Intelligence-Based Breast Cancer DiagnosisUsing Ultrasound Images and Grid-Based DeepFeature Generator.
Such techniques will be especially valuable in regions where access to specialised healthcare professionals is limited, such as developing countries.
There is a lot more work to do in this field. For one, there is currently a lack of clinical data representing all skin types, which can introduce unintended biases in the model's predictions.
Though new AI technology inspires resistance in some fiends, it is important that dermatologists accept and embrace AI as a complementary tool rather than a threat to their profession. This is how we can reduce cancer’s impact on our lives and improve outcomes for patients.
Interested in artificial intelligence?
Find out more about studying it in our guide.