Tracking Symptom Patterns and Ketogenic Therapy: Using Data and AI to Manage Mental Health

How periodic symptom analysis and a ketogenic diet are helping me manage mental illness, with the support of AI for data insights and daily keto recipes

In previous blog posts Finding Hope After Decades of Struggle, My Journey: From Mental Health Struggles to Hope Through the Power of Keto, How the Ketogenic Diet Helped Me Regain Focus and Energy, I shared my experiences with mental illness and how I’ve been using ketogenic therapy with promising results so far. Over the years, I noticed a pattern in the intensity of my symptoms, which often required me to increase and then taper off my medication—a process that typically took a couple of weeks.

To track these symptom relapses, I started logging the date, symptom intensity, and medication dosage in a Google Sheet, and I’ve been doing this consistently. I was curious to identify the precise periodicity of my relapses, so I prepared the data, filled in the missing daily entries, and generated a CSV file. Then, I asked ChatGPT to help me write a Python script to analyze the power spectrum of the data:

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from scipy.fft import fft, fftfreq
from scipy.signal import periodogram, spectrogram

# Load the data
data = pd.read_csv('medical_journal.csv', parse_dates=['Data'])
data.set_index('Data', inplace=True)

# Fill missing values with 0 and resample to daily frequency
data = data.fillna(0).resample('D').asfreq().fillna(0)

# Extract the 'Symptoms' column
symptoms = data['Symptoms'].values

# Compute FFT
n = len(symptoms)
fft_values = fft(symptoms)
fft_freq = fftfreq(n, d=1)  # Assuming daily sampling

# Compute the power spectrum
power_spectrum = np.abs(fft_values)**2

def plot_power_spectrum(freq, power, max_freq=0.5):
    plt.figure(figsize=(12, 6))
    plt.plot(freq[freq > 0], power[freq > 0])
    plt.xlabel('Frequency (cycles/day)')
    plt.ylabel('Power')
    plt.title('Power Spectrum of Symptoms Data')
    plt.xlim(0, max_freq)
    plt.grid(True)
    plt.savefig('power_spectrum.png')
    plt.close()

# Plot the power spectrum
plot_power_spectrum(fft_freq, power_spectrum)

def find_significant_frequencies(freq, power, n_peaks=5):
    positive_freq = freq > 0
    sorted_indices = np.argsort(power[positive_freq])[::-1]
    top_indices = sorted_indices[:n_peaks]
    significant_freq = freq[positive_freq][top_indices]
    return significant_freq

significant_freq = find_significant_frequencies(fft_freq, power_spectrum)
print("Most significant frequencies (cycles/day):", significant_freq)
print("Corresponding periods (days):", 1/significant_freq)

def plot_periodogram(data, fs=1, max_freq=0.5):
    f, Pxx = periodogram(data, fs)
    plt.figure(figsize=(12, 6))
    plt.semilogy(f, Pxx)
    plt.xlabel('Frequency (cycles/day)')
    plt.ylabel('Power Spectral Density')
    plt.title('Periodogram of Symptoms Data')
    plt.xlim(0, max_freq)
    plt.grid(True)
    plt.savefig('periodogram.png')
    plt.close()

plot_periodogram(symptoms)

def plot_spectrogram(data, fs=1, nperseg=30):
    f, t, Sxx = spectrogram(data, fs, nperseg=nperseg)
    plt.figure(figsize=(12, 6))
    plt.pcolormesh(t, f, np.log10(Sxx), shading='gouraud')
    plt.ylabel('Frequency (cycles/day)')
    plt.xlabel('Time (days)')
    plt.title('Spectrogram of Symptoms Data')
    plt.colorbar(label='Log Power')
    plt.savefig('spectrogram.png')
    plt.close()

plot_spectrogram(symptoms)

# Save significant frequencies to a file
np.savetxt('significant_frequencies.txt', significant_freq, fmt='%f')

Running the script resulted in the following significant frequencies:

0.014660

0.034555

0.015707

0.001047

0.019895

By taking the inverse of the main frequency, I calculated a period of 68.21 days, which aligns well with my empirical observations of my relapse cycle.

I’m now using this value as a reference to track the effectiveness of the ketogenic therapy I’m following. It’s still early, but my hope is that future relapses will become less frequent—or even stop—while I maintain a state of ketosis.

Additionally, ChatGPT has been helpful in generating a new keto recipe for me every day, ensuring I stay on track with the right macronutrient proportions to maintain ketosis.

Thank you for reading, and I hope you enjoy the rest of your day.