AI and Multispectral Photoacoustic Imaging to Diagnose Thyro
Researchers are reporting on having developed a technique for minimally invasive diagnosis of thyroid cancer. The method combines multispectral photoacoustic imaging and machine learning and is conceived as a replacement for invasive and occasionally inaccurate fine-needle aspiration biopsies. The new technique involves analyzing the unique photoacoustic signatures of malignant thyroid nodules and then training the system to recognize them.

The vast majority of thyroid nodules are benign, but as 5-10% are not, it is advisable to get them checked out. At present, clinicians use a fine needle to obtain a biopsy from the nodule. This is invasive, and in approximately 20% of cases will need to be repeated, as the results are unreliable.

As an alternative, these researchers have developed a non-invasive technique, based on the photoacoustic effect, where light absorbed by a sample produces sound waves. The researchers developed their system on the premise that the oxygen saturation in malignant thyroid nodules is lower than that in benign nodules, and that photoacoustic imaging could help to detect these differences non-invasively.

In practice, this would mean illuminating a patient’s thyroid nodule using a laser and then recording the resulting ultrasound signal. The researchers analyzed such data from patients with malignant and benign thyroid nodules using machine learning. This led them to be able to train their system to identify the malignant nodules.

So far, they have been able to detect malignant nodules with a reasonable level of accuracy, with a reported sensitivity of 83% and a specificity of 93% when the machine learning approach is combined with conventional ultrasound assessments.