Abstract P6-02-12: Artificial Intelligence over thermal images for radiation-free breast cancer screening

Abstract

Introduction: Breast cancer is the largest cause of cancer deaths in women today. NIRAMAI has developed a novel solution for detecting early stage breast cancer in women of all age groups. It is low cost, non-contact and portable solution. This radiation-free solution also works on dense breasts and hence is applicable beyond developing countries. The core of the solution is a Computer Aided Diagnostics engine called Thermalytix, which uses Artificial Intelligence algorithms on high resolution thermal images. In this paper, we present a comparative analysis of Thermalytix solution with Mammography, the standard screening modality, in a retrospective trial that was conducted across 3 reputed cancer hospitals in India. Aim: To compare the sensitivity, specificity, NPV and PPV of Thermalytix with Mammography. Methods: A multisite comparative study was performed on 194 patients across 3 reputed cancer hospitals in India with informed consent from subjects and ethics committee approval of respective hospitals. Every person who was going for a mammography test was made to undergo the non-invasive Thermalytix test prior to mammography examination. As per standard of care in India, all women who had a suspicious lesion in Mammo was sent to Ultrasound and then biopsy. 93 of the 194 subjects enrolled were found to be malignant by this standard procedure (9 of them did not have a biopsy report). Niramai Thermaytix test gave the automated reports detecting patients with suspected malignancy and those results were compared with ground truth derived from mammography, sono-mammography and biopsy. In a similar manner, standalone mammography observation was compared with the above ground truth to compare the two modalities as individual modalities. Results: Out of the 194 subjects, 93 positive cases and Niramai accurately detected 91 cases and called out one as boundary case. Sensitivity of Thermalytix was 98% and NPV was 97%, while sensitivity of Mammography was 94% with an NPV of 94%. NIRAMAI Thermalytix detected 4 more malignant patients compared to Mammography. On the other hand, mammography scored over Thermalytix in PPV and Specificity by 10%. These results are tabulated at Comparing NIRAMAI Thermalytix with Mammography NIRAMAIMammographySensitivity98%94%Specificity76%89%PPV79%89%NPV97%94%Accuracy87%91%Number of subjects = 194 high risk patients Among the 194 patients, 39 women were found to have dense breasts and 19 with heterogenous dense breasts – on whom ultrasound and elastography was performed by the Radiologist to correlate with mammography results and make the final observations. Niramai test alone was effective in these patients. The results show that Thermalytix is an emerging, radiation-free diagnostic modality that has comparable accuracy to Mammography. We conclude that NIRAMAI Thermalytix can be complementary modality to mammography as it works well on women of all age groups, including women with dense breasts. Its higher sensitivity and NPV adds to its potential to be the first screening test for early detection of breast cancer. Citation Format: Manjunath G, Sudhakar S, Kakileti S, Madhu H, Singh A. Artificial Intelligence over thermal images for radiation-free breast cancer screening [abstract]. In: Proceedings of the 2018 San Antonio Breast Cancer Symposium; 2018 Dec 4-8; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2019;79(4 Suppl):Abstract nr P6-02-12.

Personalized risk prediction for breast cancer pre-screening using artificial intelligence and thermal radiomics

“…Our own earlier research on using machine learning for breast thermography has shown promising accuracy results [20][21][22][23]. In [20], we described the effectiveness of medically interpretable imaging features obtained from the abnormal thermal patterns to get high levels of sensitivity and specificity of interpretation.…”

“…In [21], we proposed the use of shape-based and temperature-based image processing features to detect the vasculatures from thermal images. These algorithms have shown comparable and sometimes better results than standard of care [22,[21][22][23].…”