Imagine a future where a computer can assist doctors in diagnosing complex diseases with greater speed and accuracy. Researchers at Universiti Teknologi Malaysia (UTM) are one step closer to making this a reality, thanks to their work using deep learning (DL) to distinguish early-stage mycosis fungoides (MF), a type of cutaneous lymphoma, from benign inflammatory skin conditions.
Diagnosing early-stage MF is notoriously difficult because its symptoms and microscopic features often resemble those of common skin conditions like eczema or psoriasis. This can lead to delays in diagnosis and treatment. The UTM team recognized that deep learning, a type of artificial intelligence that has shown remarkable success in image recognition, could potentially provide a valuable tool for pathologists. Past research has demonstrated deep learning’s utility in classifying various cancers, the team decided to explore its application in diagnosing cutaneous lymphomas.
The researchers developed and tested DL models using a unique dataset of 924 skin biopsy images, including samples from 233 patients with early-stage MF and 353 patients with benign inflammatory dermatoses. All MF diagnoses were confirmed through clinicopathological correlation, ensuring accuracy. They then compared the performance of their AI models against three expert pathologists. The AI performed best using images magnified 200 times, achieving an average accuracy of 76.2%, nearly matching the 77.7% accuracy of the human experts. Furthermore, in the majority of cases, the AI’s ‘attention heatmaps’ – highlighting the areas it focused on – aligned with the regions of interest identified by the pathologists.
This study demonstrates the potential of using deep learning to improve the accuracy and speed of diagnosing early-stage MF. While the technology is not yet ready for clinical use, the findings are a significant step forward. The research team acknowledges that achieving clinical-grade performance will require larger, more diverse datasets from multiple institutions, as well as further refinements to the AI models, such as incorporating clinical data.
The next steps involve expanding the dataset and exploring multimodal deep learning techniques to further enhance the accuracy and reliability of the AI system. Ultimately, this research could lead to earlier and more accurate diagnoses of MF, improving patient outcomes and quality of life. This study highlights the potential of artificial intelligence to revolutionize medical diagnostics and improve healthcare for all. https://doi.org/10.52866/2788-7421.1251
