Background: In emerging markets, a routine mammogram is not feasible due to cost, accessibility and lack of skilled technicians. It is also not effective in women with dense breasts. We explore an alternative modality based on Artificial Intelligence over thermal images for early detection of breast cancer. Infrared imaging is a non-contact, radiation-free technique that detects breast abnormalities based on thermal changes in the body seen due to increased vascularity and metabolic activity of the tumour. Thermography is only approved as an adjunct modality by US FDA, as manual interpretation of thermal images is error prone and subjective. We have developed an AI- based software (called Thermalytix) for automatic analysis of Thermal images using machine learning to extract and analyse abnormal patterns based on thermal distribution, asymmetry and vascular structure; to identify malignancy in symptomatic and asymptomatic subjects. Methods: We compare performance of Thermalytix with Mammography on a total of 769 subjects who walked into 3 sites for mammography screening. Subjects were sent for NIRAMAI screening prior to other tests (mammography or sonomammography). Pregnant women, lactating mothers and cancer survivors were excluded from the study. 534 subjects had either lump or nipple discharge as a symptom and 235 subjects did not have any symptoms. Mammography was conducted on 407 subjects and rest went through sonomammography as per the discretion of radiologist. Results: Sensitivity of Thermalytix was 92% in symptomatic women with a specificity of 71.9%; while 100% sensitivity and 92.9% specificity was found in women without any symptoms. Mammography, on the other hand, presented 91% sensitivity and 70.7% specificity in symptomatic women. In asymptomatic women, mammography however showed 67% sensitivity and 98.9% specificity. Conclusions: Thermalytix was found to be a promising screening and diagnostic tool while mammography worked better as a diagnostic tool.