Challenges of AI driven diagnosis of chest X-rays transmitted through smart phones: a case study in COVID-19.

Healthcare delivery during the initial days of outbreak of COVID-19 pandemic was badly impacted due to large number of severely infected patients posing an unprecedented global challenge. Although the importance of Chest X-rays (CXRs) in meeting this challenge has now been widely recognized, speedy diagnosis of CXRs remains an outstanding challenge because of fewer Radiologists. The exponential increase in Smart Phone ownership globally, including LMICs, provides an opportunity for exploring AI-driven diagnostic tools when provided with large volumes of CXRs transmitted through Smart Phones. However, the challenges associated with such systems have not been studied to the best of our knowledge. In this paper, we show that the predictions of AI-driven models on CXR images transmitted through Smart Phones via applications, such as WhatsApp, suffer both in terms of Predictability and Explainability, two key aspects of any automated Medical Diagnosis system. We find that several existing Deep learning based models exhibit prediction instability-disagreement between the prediction outcome of the original image and the transmitted image. Concomitantly we find that the explainability of the models deteriorate substantially, prediction on the transmitted CXR is often driven by features present outside the lung region, clearly a manifestation of Spurious Correlations. Our study reveals that there is significant compression of high-resolution CXR images, sometimes as high as 95%, and this could be the reason behind these two problems. Apart from demonstrating these problems, our main contribution is to show that Multi-Task learning (MTL) can serve as an effective bulwark against the aforementioned problems. We show that MTL models exhibit substantially more robustness, 40% over existing baselines. Explainability of such models, when measured by a saliency score dependent on out-of-lung features, also show a 35% improvement. The study is conducted on WaCXR dataset, a curated dataset of 6562 image pairs corresponding to original uncompressed and WhatsApp compressed CXR images. Keeping in mind that there are no previous datasets to study such problems, we open-source this data along with all implementations.

Evaluation of Non-Invasive Thermal Imaging for Detection of Viability of Onchocerciasis Worms.

Onchocerciasis is causing blindness in over half a million people in the world today. Drug development for the disease is crippled as there is no way of measuring effectiveness of the drug without an invasive procedure. Drug efficacy measurement through assessment of viability of onchocerca worms requires the patients to undergo nodulectomy which is invasive, expensive, time-consuming, skill-dependent, infrastructure dependent and lengthy process. In this paper, we discuss the first-ever study that proposes use of machine learning over thermal imaging to non-invasively and accurately predict the viability of worms. The key contributions of the paper are (i) a unique thermal imaging protocol along with pre-processing steps such as alignment, registration and segmentation to extract interpretable features (ii) extraction of relevant semantic features (iii) development of accurate classifiers for detecting the existence of viable worms in a nodule. When tested on a prospective test data of 30 participants with 48 palpable nodules, we achieved an Area Under the Curve (AUC) of 0.85. Clinical Relevance- This is the first ever research effort of using thermal imaging in the assessment of viability of onchocerca worms and it resulted in a very high specificity>95% which makes it a promising modality to pursue further.

Automated vascular analysis of breast thermograms with interpretable features.

Purpose: Vascular changes are observed from initial stages of breast cancer, and monitoring of vessel structures helps in early detection of malignancies. In recent years, thermal imaging is being evaluated as a low-cost imaging modality to visualize and analyze early vascularity. However, visual inspection of thermal vascularity is challenging and subjective. Therefore, there is a need for automated techniques to assist physicians in visualization and interpretation of vascularity by marking the vessel structures and by providing quantified qualitative parameters that helps in malignancy classification Approach: In the literature, there are very few approaches for vascular analysis and classification of breast thermal images using interpretable vascular features. One major challenge is the automated detection of breast vascularity due to diffused vessel boundaries. We first propose a deep learning-based semantic segmentation approach that generates heatmaps of vessel structures from two-dimensional breast thermal images for quantitative assessment of breast vascularity. Second, we extract interpretable vascular parameters and propose a classifier to predict likelihood of breast cancer purely from the extracted vascular parameters. Results: The results of the cancer classifier were validated using an independent clinical dataset consisting of 258 participants. The results were encouraging as the proposed approach segmented vessels well and gave a good classification performance with area under receiver operating characteristic curve of 0.85 with the proposed vascularity parameters. Conclusions: The detected vasculature and its associated high classification performance show the utility of the proposed approach in interpretation of breast vascularity.

A prospective evaluation of breast thermography enhanced by a novel machine learning technique for screening breast abnormalities in a general population of women presenting to a secondary care hospital.

Objective: Artificial intelligence-enhanced breast thermography is being evaluated as an ancillary modality in the evaluation of breast disease. The objective of this study was to evaluate the clinical performance of Thermalytix, a CE-marked, AI-based thermal imaging test, with respect to conventional mammography. Methods: A prospective, comparative study performed between 15 December 2018 and 06 January 2020 evaluated the performance of Thermalytix in 459 women with both dense and nondense breast tissue. Both symptomatic and asymptomatic women, aged 30-80 years, presenting to the hospital underwent Thermalytix followed by 2-D mammography and appropriate confirmatory investigations to confirm malignancy. The radiologist interpreting the mammograms and the technician using the Thermalytix tool were blinded to the others' findings. The statistical analysis was performed by a third party. Results: A total of 687 women were recruited, of whom 459 fulfilled the inclusion criteria. Twenty-one malignancies were detected (21/459, 4.6%). The overall sensitivity of Thermalytix was 95.24% (95% CI, 76.18-99.88), and the specificity was 88.58% (95% CI, 85.23-91.41). In women with dense breasts (n = 168, 36.6%), the sensitivity was 100% (95% CI, 69.15-100), and the specificity was 81.65% (95% CI, 74.72-87.35). Among these 168 women, 37 women (22%) were reported as BI-RADS 0 on mammography; in this subset, the sensitivity of Thermalytix was 100% (95% CI, 69.15-100), and the specificity was 77.22% (95% CI, 69.88-83.50). Conclusion: Thermalytix showed acceptable sensitivity and specificity with respect to mammography in the overall patient population. Thermalytix outperformed mammography in women with dense breasts and those reported as BI-RADS 0.

Observational Study to Evaluate the Clinical Efficacy of Thermalytix for Detecting Breast Cancer in Symptomatic and Asymptomatic Women.

PURPOSE: To evaluate the sensitivity and specificity of Thermalytix, an artificial intelligence-based computer-aided diagnostics (CADx) engine, to detect breast malignancy by comparing the CADx output with the final diagnosis derived using standard screening modalities. METHODS: This multisite observational study included 470 symptomatic and asymptomatic women who presented for a breast health checkup in two centers. Among them, 238 women had symptoms such as breast lump, nipple discharge, or breast pain, and the rest were asymptomatic. All participants underwent a Thermalytix test and one or more standard-of-care tests for breast cancer screening, as recommended by the radiologists. Results from Thermalytix and standard modalities were obtained independently in a blinded fashion for comparison. The ground truth used for analysis (normal or malignant) was the final impression of an expert clinician based on the symptoms and the available reports of standard modalities (mammography, ultrasonography, elastography, biopsy, fine-needle aspiration cytology, and so on). RESULTS: For the 470 women, Thermalytix resulted in a sensitivity of 91.02% (symptomatic, 89.85%; asymptomatic, 100%) and specificity of 82.39% (symptomatic, 69.04%; asymptomatic, 92.41%) in detection of breast malignancy. Thermalytix showed an overall area under the curve (AUC) of 0.90, with an AUC of 0.82 for symptomatic and 0.98 for asymptomatic women. CONCLUSION: High sensitivity and high AUC of Thermalytix in women of all age groups demonstrates the efficacy of the tool for breast cancer screening. Thermalytix, with its automated scoring and image annotations of potential malignancies and vascularity, can assist the clinician in better decision making and improve quality of care in an affordable and radiation-free manner. Thus, we believe Thermalytix is poised to be a promising modality for breast cancer screening.

Robust Estimation of Breast Cancer Incidence Risk in Presence of Incomplete or Inaccurate Information.

PURPOSE: To evaluate the robustness of multiple machine learning classifiers for breast cancer risk estimation in the presence of incomplete or inaccurate information. DATA AND METHODS: Open data for this study was obtained from the BCSC Data Resource (http://breastscreening.cancer.gov/). We conducted two ablation-type experiments to compare the robustness of different classifiers where we randomly switched known information to missing with a missing probability of pm in one experiment, and randomly corrupted the existing information with a probability of pc in another experiment. We considered three prominent machine-learning classifiers such as Logistic regression (LR), Random Forests (RF) and a custom Neural Network (NN) architecture and compared their degradation of discrimination performance as a function of increasing probability of missing or inaccurate data. RESULTS: LR, RF and custom NN resulted in an Area Under Curve (AUC) of 0.645, 0.643 and 0.649, respectively, on a test set with 500,000 total observations. When we manipulated the data by varying probabilities pm and pc from 0 to 1, NN resulted in better performance in terms of AUC compared to RF and LR as long as less than half the data was missing/inaccurate (that is, for values of pm < 0.5 and pc < 0.5). However, for missing (pm) or corruption (pc) probabilities above 0.5, LR gave similar performance as the custom NN. RF resulted in overall poorer performance when the data had additional missing or incorrect entries. CONCLUSION: In cases where the input information is missing or inaccurate, our experiments show that the proposed custom NN provides reliable risk estimates in medical datasets like BCSC. These results are particularly important in health care applications where not every attribute of the individual participant might be available.
.

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

MOTIVATION: Breast cancer is the leading cause of cancer deaths among women today. Survival rates in developing countries are around 50%-60% due to late detection. A personalized, accurate risk scoring method can help in targeting the right population for follow-up tests and enables early detection of breast abnormalities. Most of the available risk assessment tools use generic and weakly correlated features like age, weight, height etc. While a personalized risk scoring from screening modalities such as mammography and ultrasound could be helpful, these tests are limited to very few metropolitan hospitals in developing countries due to high capital cost, operational expenses and interpretation expertise needed for a large screening population. METHODS: We propose and analyze a new personalized risk framework called Thermalytix Risk Score (TRS) to identify a high-risk target population for regular screening and enable early stage breast cancer detection at scale. This technique uses Artificial Intelligence (AI) over thermal images to automatically generate a breast health risk score. This risk score is mainly derived from two sub-scores namely, vascular score and hotspot score. A hotspot score signifies the abnormality seen from irregular asymmetric heat patterns seen on the skin surface, whereas vascular score predicts the presence of asymmetric vascular activity. These scores are generated using machine learning algorithms over medically interpretable parameters that describes the metabolic activity inside the breast tissue and indicate the presence of a possible malignancy even in asymptomatic women. RESULTS: The proposed personalized risk score was tested on 769 subjects in four breast cancer screening facilities. The subjects' age ranged from 18 to 82 years with a median of around 45 years. Out of the 769 subjects, 185 subjects were diagnosed with a breast malignancy by an expert radiologist after mammography, ultrasound and/or histopathology. Our personalized AI based risk score achieved an area under the receiver-operator curve (AUC) of 0.89 when compared to an age normalized risk score that showed an AUC of 0.68. We also found that if the computed risk score is used to place individuals into four risk groups, the likelihood of malignancy also increases monotonically with the risk grouping level. CONCLUSION: The proposed AI based personalized risk score uses breast thermal image patterns for risk computation and compares favorably to other generic risk estimation approaches. The proposed risk framework solution is automated, affordable, non-invasive, non-contact and radiation free and works for a wide age range of women from 18 to 82 years, including young women with dense breasts. The proposed score might be further used to assign subjects into one of the four risk groups and provide guidance on the periodicity of screening needed. In addition, the automatically annotated thermal images localizes the potential abnormal regions and might empower the physician to create a better personalized care.

Cascaded CNN for View Independent Breast Segmentation in Thermal Images.

Breast Cancer is the leading cause of cancer deaths in women today. Use of thermal imaging for early stage breast cancer screening is gaining more adoption in recent times and automated analysis of these thermal images with computer aided diagnosis is the key to maintain objectivity in assessment and improve quality of diagnosis. One of the main challenges in automated breast thermography is accurate segmentation of breast region robust to technician errors in image capture - such as view, distance from imaging device, position, etc. Existing algorithms for segmentation are mostly based on heuristic rules and are highly dependent upon the image capture correctness. We propose a cascaded CNN architecture to perform accurate segmentation robust to subject views and capture errors. The proposed approach can detect breasts region independent of the image capture and view angle, enabling automated image and video analysis. We also detailed and compared our algorithm with a multi-view heuristics-based segmentation method. Our proposed technique resulted a dice index of 0.92 when compared with expert segmentation on a test set comprising of 900 images collected from 150 subjects at five different view angles.