for any thermograms there is an excel file representing pixels' temperature. I can segment the images automatically, but for further image analysis expert's validation is required
Hi,
do you have a good gold standard classification? I mean: do you know the exact diagnosis of each of the subject?
Best regards
AM
Dear Arcangelo Merla,
No, I don't have the exact diagnosis for each image.
If I had them I wouldn't need expert evaluation.
Thank you for the question. The main issue you have is the lack of ground truth for your data, as pointed out by Dr Merla. You would be best served by trying to find this for your own cases. There must be records somewhere so that you can check against the pathology databases.
You are also confusing terms regarding infra red imaging, and thermography. Thermography is different from thermal imaging, and has been investigated over several decades and is extremely controversial due to lack of evidence of specificity.
Near Infra-Red (NIR) imaging by use of IR cameras in conjunction with a phosphorescent marker you can visualize cancer calls individually both in vitro and in vivo. There is much excellent research in many types of cancer using these techniques, which work especially well for lesions in relatively superficial organs like the breast.
My recomendation would be to obtain a database of any kind of imaging of the breast that has ground truth pathology available. That would be the best data set to do research with.
Hope that helps. Chris
Dear Professor Chris,
Thank you for your thorough answer.
What is in my mind, is to establish a somehow automatic method for discrimination between for example normal and abnormal or benign or malignant images.
So the subjective evaluation made by thermology expert based on thermal images can be my ground truth.
Nasser
As you do not know which of the cases had cancer, which had abnormal findings (which could potentially produce false positive or confounding results) and which were normal it makes no difference to the data you will get from an 'expert' reading, so your conclusion would be worthless.
Compare this to looking at a mammogram, if you have a spiculate mass it is likey to be read as highly suspicious of malignancy BI-RADS 5. However, after biopsy, ground truth (pathology) is established which may show fat necrosis, which simulates the gross anatomical features of a cancer.
In other words, the expert reader was wrong according to ground truth, despite the probability of a BI-RADS 5 lesion being >95% likely to be malignant.
chris
Dear Nasser,
I recomend you read these papers:
"Non-invasive estimation of the metabolic heat production of breast tumors using digital infrared imaging" by Gonzalez
And
“Infrared imaging of the breast: initial reappraisal using high-resolution digital technology in 100 successive cases of stage I and II breast cancer” by Keyserlingk
In both articles, the interpretation of the images was done using a thermal score derived from the Ville Marie Infrared grading scale.
Best regards
Hi,
I'll be glad to help you diagnosing them through classification
please feel free to contact.
Ameen Mohammed Abd Al-salam Selami Thank you very much. But I need clinical evaluation by medical persons rather than engineering classification.
Dear Nasser,
If you want to send me these pictures. I try to classify them. But you should know that this will be my first experience in this area. However, with my 15 years of clinical experience, I can say that I can distinguish between benign and malignant, at least I will try.
Murat