Dermatology researchers in the US have developed an artificial intelligence program that uses histology samples to identify which patients with melanoma will respond to immunotherapy.
They say the system offers a better performance than genomic biomarkers such as PD-L1, which may be modified during the course of immune checkpoint inhibitor therapy.
Instead, their machine learning system applied deep learning on haematoxylin and eosin–stained (H&E) slides of metastatic lymph node and subcutaneous tissue, the usual visual microscopic method of for diagnosing melanoma and staging disease severity.
Writing in Clinical Cancer Research, clinicians from NYU Grossman School of Medicine and Perlmutter Cancer Center said they collected 302 images of tumour tissue samples from 121 men and women treated for metastatic melanoma with immune checkpoint inhibitors at NYU Langone hospitals
They divided the digital images from slide scanners into 1.2 million portions of pixels that were fed into the computer along with factors such as the severity of the disease, which kind of immunotherapy regimen was used, and whether a patient responded to the treatment. The study investigators repeated this process with 40 slides from 30 similar patients as a validation model.
In their analysis, the multivariable classifier predicted response with an AUC 0.800 on histology images, and accurately stratified patients into high versus low risk for disease progression.
Melanoma patients classified as high risk for progression had significantly worse progression free survival (PFS) than those classified as low risk (P = 0.02 for the Aperio AT2 slide scanner; P = 0.03 for the Leica SCN400).
“Our proposed model overcomes the limitations of the temporal and spatial heterogeneity that impede the performance of PD-L1 as well as the resource scarcity that precludes using RNA sequencing, while simultaneously maintaining its validity across multiple slide scanners,” the researchers wrote.
“Our approach is time efficient, reproducible, and requires minimal resource allocation, thus overcoming multiple common barriers to generalisability for contemporary biomarkers.”
With further optimisation of the model using larger datasets, they believed the model could be used in clinical practice to help identify melanoma patients who are at high versus low risk for progression through immunotherapy.
The researchers also noted that aside from the computer needed to run the program, all of the materials and information used in the technique were already a standard part of cancer management that most clinics use.
And they argued that the AI method was more streamlined than current predictive tools such as analysing stool samples or genetic information, which promises to reduce treatment costs and speed up patient wait times.
“Even the smallest cancer centre could potentially send the data off to a lab with this program for swift analysis,” said study senior author Dr Iman Osman, Professor of Dermatology at NYU Langone and its Perlmutter Cancer Center.