Non-tuberculous mycobacteria (NTM) are a type of mycobacteria that cause pulmonary disease in humans. Normally, the human immune system can maintain a balance with the invading NTM bacteria, which is called “Colonization”. When the immune system becomes weak, NTM will cause “Disease” and bring about further injury to the human body. Distinguishing between Colonization and Disease requires several doctors to agree on a final diagnosis by analyzing computed tomography (CT) scans and bacterial samples taken from patient’s sputum, which would consume much time and medical resources. Therefore, finding an efficient and effective way to identify NTM may be helpful for doctors.
In this work, we propose a mixture model of 3D convolutional neural network (3D CNN) and multilayer perceptron (MLP) called NTM-Net, with our newly designed labelling method called Weighted Labelling which is for binary classification task with non-binary training data like NTM status, which can automatically identify a suspected NTM patient with his/her CT scans and clinical data. Experiments show that our hybrid model with weighted labelling comprehensively
outperforms models based on CNN or MLP alone, or without our proposed labelling method.
We also develop a Human-in-the-Loop (HITL) process to help doctors re-evaluate cases without consensus by interacting with NTM-Net through the predictions and attention heat maps it generates for several rounds. Our experiments show that applying Human-in-the-Loop (HITL) process can slightly enhance the performance of NTM-Net. The analysis of re-evaluation of NTM status during the Human-in-the-Loop (HITL) process also shows that doctors are easier to have consensus when re-evaluating cases which are unconsensus in previous round after referencing predictions and attention heat maps from NTM-Net. Besides, feedback of questionnaire from doctors shows that prediction of NTM-Net is reasonable and useful for re-evaluation on cases without consensus for doctors during the Human-in-the-Loop (HITL) process.