Krithika Venkataramani

Extraction of medically interpretable features for classification of malignancy in breast thermography

Thermography, with high-resolution cameras, is being re-investigated as a possible breast cancer screening imaging modality, as it does not have the harmful radiation effects of mammography. This paper focuses on automatic extraction of medically interpretable non-vascular thermal features. We design these features to differentiate malignancy from different non-malignancy conditions, including hormone sensitive tissues and certain benign conditions, which have an increased thermal response. These features increase the specificity for breast cancer screening, which had been a long known problem in thermographic screening, while retaining high sensitivity. These features are also agnostic to different cameras and resolutions (up to an extent). On a dataset of around 78 subjects with cancer and 187 subjects without cancer, that have some benign diseases and conditions with thermal responses, we are able to get around 99% specificity while having 100% sensitivity. This indicates a potential break-through in thermographic screening for breast cancer. This shows promise for undertaking a comparison to mammography with larger numbers of subjects with more data variations.Thermography, with high-resolution cameras, is being re-investigated as a possible breast cancer screening imaging modality, as it does not have the harmful radiation effects of mammography. This paper focuses on automatic extraction of medically interpretable non-vascular thermal features. We design these features to differentiate malignancy from different non-malignancy conditions, including hormone sensitive tissues and certain benign conditions, which have an increased thermal response. These features increase the specificity for breast cancer screening, which had been a long known problem in thermographic screening, while retaining high sensitivity. These features are also agnostic to different cameras and resolutions (up to an extent). On a dataset of around 78 subjects with cancer and 187 subjects without cancer, that have some benign diseases and conditions with thermal responses, we are able to get around 99% specificity while having 100% sensitivity. This indicates a potential break-through in thermographic screening for breast cancer. This shows promise for undertaking a comparison to mammography with larger numbers of subjects with more data variations.

Semi-automated breast cancer tumor detection with thermographic video imaging.

Screening for breast cancer enables early detection by which curative treatment can be possible. While mammography is the current gold standard for screening, it has low sensitivity in younger women and its harmful X-rays in frequent screening can increase the risk of cancer. Incidence rates are rising in younger women, causing a relook at thermography for low cost and non-harmful screening. In this paper, thermography is compared to mammography correlated with sono-mammography in 65 FNAC/biopsy proven cancer subjects in India. Thermography is comparable to mammography correlated with sono-mammography, having 94% and 95% sensitivity, respectively. A novel semi-automated thermographic tumor detection and location algorithm used in this paper also provides 97% sensitivity. This shows the promise of automated thermographic screening for reaching large populations in a cost effective manner in low resource settings in countries like India. Further studies in a large scale need to be done to evaluate the specificity to enable such solutions.

Automatic Determination of Hormone Receptor Status in Breast Cancer Using Thermography

Estrogren and progesterone hormone receptor status play a role in the treatment planning and prognosis of breast cancer. These are typically found after Immuno-Histo-Chemistry (IHC) analysis of the tumor tissues after surgery. Since breast cancer and hormone receptor status affect thermographic images, we attempt to estimate the hormone receptor status before surgery through non-invasive thermographic imaging. We automatically extract novel features from the thermographic images that would differentiate hormone receptor positive tumors from hormone receptor negative tumors, and classify them though machine learning. We obtained a good accuracy of 82 % and 79 % in classification of HR\(+\) and HR− tumors, respectively, on a dataset consisting of 56 subjects with breast cancer. This shows a novel application of automatic thermographic classification in breast cancer prognosis.

A Method for Classifying Cancerous and Normal Regions in Breast Thermography for Detecting Smaller Size Tumors

Breast cancer (BC) is one of the most frequently diagnosed cancers in women. In the United States, one in eight women are likely to be diagnosed with having some form of BC in her lifetime. Prevention through frequent screening can minimize the risk of mortality through BC. The ability to obtain frequent screening in low resource settings as in India is limited for relatively large populations of women. The earlier the cancer can be detected, the more likelihood that the patient responds to treatment. Accordingly, new technologies and methodologies for the detection of cancer are increasingly needed. While mammography is the current gold standard for screening, it has low sensitivity in younger women and its harmful Xrays in frequent screening can increase the risk of cancer. Incidence rates are rising in younger women, causing a relook at thermography. We propose a sparse coding-based method to jointly learn a discriminative dictionary and an optimal linear classifier for semi-automated BC detection. Training data is built from small size image patches of normal and cancerous regions. We hypothesize that the size of the detectable tumor with the proposed method could be very small. In our method, each training sample is represented as a sparse linear combination of dictionary atoms. The resulting sparse representations are used as features for the classifier. Partial proof-of-concept is shown with 96% sensitivity for sample thermal patches of malignant and normal regions. Technique is compared to mammography correlated with sonomammography in 45 FNAC/biopsy proven cancer subjects.

Towards automated selection of data fusion techniques

An investigation of multi-modal fusion schemes is done using synthetic data generation to determine how the data characteristics influence fusion. The goal is to select the best fusion scheme using data characteristics. Preliminary results are presented here that compare data concatenation to Kernel fusion in the presence of increasing dimensionality, linear/nonlinear decision boundaries and correlations between different modality features. It is found that data concatenation is better than Kernel fusion in low dimensions in general. It is also found that Kernel fusion is better than data concatenation when the optimal decision boundary is non-linear, and the dimensions are high. Correlations between modalities determine the information content, and Kernel fusion reduces the information content most when there is negative correlation between modalities. These results are applied to fingerprint live-ness detection on the ATVS database having three sensor modalities. As there are few features used per modality and the overall dimensionality is low, it is expected and confirmed that data concatenation is better than Kernel fusion.